An anti-corruption system for academic achievement verification in kazakhstani higher education using blockchain and artificial intelligence

This article focuses on developing an anti-corruption system for certifying students’ academic achievements in Kazakhstani higher education institutions by utilizing blockchain and artificial intelligence AI technologies. We specifically propose the Academic Integrity Verification System (AIVS), a revolutionary system that combines blockchain’s tamper-proof storage with AI’s anomaly detection capabilities. The system reduces major risks in traditional academic record management while ensuring transparency, precision, and proactive fraud detection. The simulation was conducted at the International Information Technology University (IITU) using Ethereum-based blockchain and AI models. In simulated testnet experiments, AIVS achieved an 85% reduction in verification time compared to traditional processes and delivered a 95% overall model accuracy in record validation. These results demonstrate the potential of blockchain and AI integration for improving efficiency and integrity in academic verification workflows. These findings demonstrate that our proposed AIVS enhances academic transparency, reduces corruption, and provides a scalable framework for secure academic record management. The proposed strategy marks a significant step forward in the governance of digital education in Kazakhstan and abroad.

1 Introduction

Integrity of academic results has arisen as a critical problem for higher education institutions around the world in recent years (Tong et al., 2022). The increased frequency of counterfeit diplomas, manipulated academic records, and illegitimate grade adjustments has significantly harmed the trust of educational credentials (Gräther et al., 2018). According to the World Education Services Report (Eaton and Carmichael, 2023), academic misconduct is now widely acknowledged as a serious threat to the integrity of education around the world. According to the findings of nationwide surveys done by Kazakhstan’s Organization for Strategic Planning and Reforms, more than 30% of students have been subjected to dishonest techniques during tests or thesis defenses. Not only do these events lower the status of educational institutions, (Davletbayeva et al,. 2025) but they also have far-reaching consequences for society by undermining actual intellectual achievements. According to current government figures, the number of academic-related corruption cases and financial losses has increased significantly in Kazakhstan (Panzabekova et al., 2024). The number of corruption cases increased from 2019 to 2023, with financial losses exceeding 1 billion tenge. The continuation of these dangerous trends into 2024 and 2025 reinforced the critical need for technologically advanced and resilient solutions, such as the proposed AIVS, to defend academic institutions’ integrity. The traditional approach to academic record administration is based on centralized databases that are vulnerable to manipulation, unauthorized access, and single points of failure (Any et al., 2024). Although these systems are adequate for basic administrative needs, they are increasingly unable to meet the sophisticated nature of modern academic fraud. The vulnerability of academic credentials is exacerbated by the fact that manual verification methods are time-consuming, costly, and prone to human mistake when compared to automated verification systems (Turkanović et al., 2018). This heightens the risk of academic credentials being compromised. As a result, there is an urgent need for technical advancements that improve the integrity, transparency, and security of academic data. This obligation stems directly from the situation. The previously described condition is directly responsible for this requirement. In this domain, blockchain technology has emerged as a disruptive solution capable of revolutionizing the landscape (Razaque et al., 2022). Blockchain technology allows for the immutable recording of academic degrees. This is used to generate a decentralized, immutable ledger. As a result, it is clear that the information cannot be changed without the consent of the network’s participants (Anwar et al., 2022; Chen et al., 2024). Blockchain technology can be used in education to verify credentials, validate certificates, authenticate enrollments, and transmit academic resources securely internationally. Global endeavors such as Blockcerts (Razaque et al., 2018), Sony (Boulet et al., 2025) Global Education (Kustandi et al., 2024), Hyland Credentials (Sharwani and Melo, 2024), and Digital Credentials (Rustemi et al., 2023) demonstrate how blockchain technology can preserve academic data. These technologies enable educational institutions, organizations, and students to authenticate credentials without relying on a central authority.

Blockchain technology assures data immutability, but it does not include capabilities for detecting fraud before records are finalized. Artificial intelligence, specifically machine learning and anomaly detection systems, has the potential to narrow this gap. Artificial intelligence can examine enormous datasets, find patterns of grade inflation, detect academic performance anomalies, and predict potential instances of misbehavior (Lu et al., 2022). The integration of artificial intelligence into academic verification systems introduces a proactive layer of fraud prevention, shifting from passive data security to active surveillance and anomaly detection (Alsulami, 2024). Despite successful pilot projects on a global scale, the use of blockchain technology and artificial intelligence in higher education remains limited, notably in Central Asian countries like Kazakhstan. The primary focus of current blockchain-based solutions is record issuance and verification, with little attention paid to fraud detection or real-time transaction monitoring (Li et al., 2022). Furthermore, the majority of activities have been concentrated on the European, North American, and East Asian regions, leaving a research and implementation gap in other regions (Kamarudin et al., 2024; Khan et al., 2021).

This study proposes the development of the AIVS (a new framework that combines blockchain’s immutable storage capabilities with AI-based anomaly detection to improve the verification of students’ academic achievements in Kazakhstani higher education institutions. This approach is intended to tackle these essential issues. AIVS is designed to protect academic integrity by proactively flagging suspicious behaviors and securely recording academic credentials, maintaining credential integrity throughout their existence. The system was constructed and evaluated in a simulated environment utilized at IITU, Kazakhstan, rather than in an actual deployment. Preliminary outcomes from these simulations indicate that the proposed method might reduce verification durations by 85% and identify issues with 95% accuracy. The results indicate that the AIVS model is resilient to cyberattacks, scalable for broader applications, and adaptable for implementation in larger national and international contexts. They ought to be regarded as outcomes from simulations that indicate potential rather than actual production-level efficacy.

Figure 1 depicts the overarching design and operating sequence of the proposed academic integrity verification system. Students submit academic records that are securely hashed and stored on the blockchain to guarantee immutability. Concurrently, AI-driven anomaly detection systems scrutinize these records to uncover possible discrepancies. Verifiers, including companies or academic organizations, may request verification, which is conducted via smart contracts and AI assessments to confirm the authenticity of records and identify anomalous patterns. This comprehensive strategy creates a strong, transparent, and intelligent framework for maintaining and validating academic integrity across all stakeholders.

Figure 1

Figure 1. General framework of the academic integrity verification system.

The main objectives and contributions of this study are as follows:

⁃ The proposed system leverages the features of recurrent neural networks, convolutional neural networks, and semantic analysis to detect inconsistencies, grade inflation patterns, and academic irregularities, enabling proactive risk scoring and administrative flagging of potential misconduct.

⁃ The authenticity and integrity of student academic records in higher education institutions are guaranteed using a novel anti-corruption system that integrates Ethereum-based blockchain smart contracts with AI-driven anomaly detection.

⁃ A SHA3-512 hashing mechanism is applied to timestamped academic records, ensuring immutability and verifiability of credentials stored on the blockchain. This helps in preventing post-recording tampering or manipulation.

⁃ The framework aligns with the legal and operational landscape of Kazakhstan’s higher education system, which offers scalability to support transparency, accountability, and public trust in credential verification.

The remainder of the paper is organized as follows: Section 2 discusses the related works. Section 3 is Research methodology. Section 4 outlines our proposed AIVS and system architecture. Section 5 describes the experimental setup and reports the results. Section 6 discusses the findings, challenges, and limitations. Finally, Section 7 concludes the entire article.

2 Related work

This section discusses the related state-of-the-art. Kazakhstanis are understandably anxious about the reliability of academic records. The widespread use of counterfeit academic credentials erodes educational institutions’ reputations and undermines public belief in authentic college degrees. According to recent polls in Kazakhstan, more than 30% of students have experienced corruption during examinations or thesis assessments. To maintain the credibility of higher education, academic record management systems must be secure and transparent. Academic qualifications should be universally recognized throughout educational institutions, emphasizing the need for technological solutions. Blockchain technology overcomes these concerns by constructing decentralized, tamper-resistant ledgers to protect sensitive data (Widayanti et al., 2021). Blockchain technology is being utilized in education to safeguard digital identities, validate credentials, authenticate degrees, and facilitate global academic content sharing (Ishkov and Krupnov, 2024). Institutions can monitor academic performance without the need for a central authority to verify that credentials are trustworthy, open, and verifiable. Blockchain is becoming used in education to improve transparency, credential verification, and record-keeping (Kerimku et al., 2023). Institutions are simultaneously maintaining academic integrity and boosting administrative efficiency, and this technological revolution is gaining global attention. These advances in artificial intelligence enable comprehensive academic data anomaly detection. Machine learning algorithms and natural language processing models can analyze large amounts of student performance data to detect abnormalities that could indicate fraud, manipulation, or academic misconduct (Nadeem et al., 2023). AI systems can proactively detect questionable actions via predictive analytics and adaptive surveillance, minimizing reliance on manual oversight while improving the overall integrity of academic evaluations (Ramasamy and Khan, 2024). Numerous global initiatives have proved the feasibility of combining blockchain and AI in educational settings (Gupta et al., 2024). The BeCertify program used blockchain technology to certify academic courses with real-time student monitoring, while BCdiploma provides GDPR-compliant verifiable diplomas for faster cross-border credential verification. Recent blockchain breakthroughs have created hybrid frameworks for credential management that optimize cost, security, and scalability (Vevera and Botezatu, 2024), while credentialing systems have evolved to improve security and usability through QR code verification and decentralized architecture (Xu, 2024). These programs have made great progress; nonetheless, obstacles remain in fully resolving real-time fraud prevention and predictive anomaly detection.

Despite these developments, modern systems primarily focus on safe credential storage and passive verification, rather than proactive detection of fraudulent activity. They value the retention of validity after issuance over real-time monitoring or predictive anomaly identification. This issue emphasizes the critical need for comprehensive solutions that protect academic data while quickly detecting academic misconduct throughout the certification process. Blockchain technology is increasingly being used in education to improve transparency, certificate verification, and record-keeping. Simultaneously, institutions are maintaining academic integrity while enhancing administrative efficiency, and this technological revolution is gaining international attention (Nadeem et al., 2023). To address this gap, this work introduces AIVS, a revolutionary architecture that blends blockchain immutability with AI-based anomaly detection and predictive risk assessment. In contrast to previous blockchain-based credentialing solutions, AIVS includes a dynamic fraud protection layer that detects irregularities before credentialing is completed. AIVS seeks to build a new baseline for academic openness, security, and digital trust, hence facilitating a resilient and progressive academic governance framework for future educational institutions. It was originally developed to solve the issues facing higher education institutions in Kazakhstan, but it is now applicable worldwide.

Table 1 provides a complete comparison of existing worldwide blockchain-based academic credential verification methods with our proposed AIVS. While various global initiatives, like Blockcerts, Sony Global Education, and BCdiploma, have demonstrated the use of blockchain for academic credential verification, these systems are primarily concerned with the secure storage and retrieval of records.

Table 1

Table 1. Comparative overview of existing blockchain-based academic verification solutions and the proposed AIVS framework.

Recent advancements in blockchain research underscore the significance of scalability and anonymity, both of which are essential for the viability of systems such as ours. Layer-2 scaling solutions have become good ways to speed up transactions and ease network congestion. Narayanan et al. (2021) (Lavin et al., 2024) provide an overview of Layer-2 protocols, such as payment channels and rollups, showing how these techniques can greatly reduce gas costs while keeping the Ethereum mainnet safe. Buterin (2021) (Rostamkolaei Motlagh et al., 2025) discusses how combining Optimistic Rollups and zk-Rollups might help achieve high throughput without losing decentralization. At the same time, privacy-preserving methods like zero-knowledge proofs (zk-proofs) are being used more and more to make decentralized apps more secure and trustworthy. A recent survey by Lavin et al. (2024) (Fatima and Senthilkumar, 2025) gives a general picture of ZK-proof applications, such as blockchain privacy, scaling (e.g., through zk-Rollups), storage, and interoperability. It also talks about the supporting infrastructure, such as zero-knowledge virtual machines, domain-specific languages, and proof frameworks. Rostamkolaei Motlagh et al. (2025) just published a paper that evaluates the best ZKP protocols (zk-SNARKs, zk-STARKs, Bulletproofs) in terms of proof size, proof production and verification costs, and how well they operate in resource-limited or decentralized settings. Our proposed system is now working in an experimental context, where speed and accuracy of verification are more important than in these other approaches. But combining the best Layer-2 scaling approaches (such as zk-Rollups) with privacy-preserving proofs is still a good way to go for getting real-world deployment on a national scale. Furthermore, the proposed AIVS platform not only ensures tamper-proof credential storage but it also includes real-time fraud detection via AI-based anomaly analysis. This proactive technique permits the early detection of suspect academic behavior, surpassing the passive verification capabilities of previous technologies. Furthermore, AIVS provides a scalable, resource-efficient design that is geared precisely to the issues faced by Central Asian higher education systems, where academic corruption remains a major concern. This novel method establishes AIVS as a transformative model for maintaining academic integrity and openness in both national and international educational institutions.

3 Research methodology

The proposed AIVS architecture is based on three modules: blockchain technology, smart contracts, and artificial intelligence. These three pillars were included in the platform. Throughout the academic data certification process, each component played an important role in assuring openness, security, and the detection of anomalies in real time. This was accomplished by each component separately. One of the most essential features of the AIVS’ design is the blockchain technology that serves as its foundation. The distributed, tamper-proof ledger it provides is used to securely register academic records in the absence of centralized authority. Every academic transaction, such as grade submission or diploma issuing, uses SHA-512 to perform cryptographic hashing, with the resulting hash recorded on the Ethereum blockchain technology. This ensures the data’s integrity. This makes it difficult to make modifications to the data after it has been uploaded because doing so will result in discovery. For example, when a student finishes a course, the system immediately hashes the information about their grade and publishes it to the blockchain via a smart contract. The blockchain makes this transaction possible. Any subsequent attempts to change this record would result in a hash mismatch, informing the system that the record exists. This system would then be able to take the necessary action. AIVS reduces the need for centralized databases by leveraging blockchain technology. This reduces AIVS’s reliance on centralized databases, ensuring academic integrity even in the face of challenges from within or outside the university.

The use of smart contracts improves the security and automation of the recording and verification procedures carried out within AIVS. These contracts, developed in Solidity and deployed on the Ethereum blockchain, define the logic for entering academic records, confirming their legitimacy, and maintaining access privileges. These contracts were designed to be compatible with the Ethereum blockchain. The smart contract will automatically verify if the submission meets specific criteria (for example, matching student IDs and course codes), and it will only record the hash on the blockchain if the criteria are met. For example, when a teacher submits final grades for a course, the smart contract determines whether the submission meets these requirements. Similarly, employers or other external verifiers can trigger a smart contract mechanism that certifies a graduate’s diploma by comparing a given hash to the original recorded on the blockchain. This allows you to check the authenticity of the diploma. This automatic validation makes the process more visible, efficient, and secure by lowering the need for manual involvement and the risk of human manipulation or error. Furthermore, it reduces the likelihood of humans making mistakes or being deceived.

The artificial intelligence modules that comprise AIVS’s intelligent layer are responsible for identifying suspicious trends that may indicate academic fraud. These modules are charged with the task of discovering such patterns. The artificial intelligence subsystem continuously monitors student academic achievement using machine learning algorithms created in TensorFlow and Keras. This is done to identify any abnormalities in student academic performance. This category of anomalies includes inconsistencies in writing style across tasks, unusual submission timeframes, and rapid grade inflation. A recurrent neural network (RNN), for example, can monitor successive grades across numerous semesters and identify cases when performance unexpectedly exceeds statistical norms without prior signals. This is achievable because RNNs can detect and classify comparable events. Furthermore, natural language processing (NLP) models are used to examine written thesis submissions for any changes in linguistic style that could indicate plagiarism or ghostwriting. This is being done to guarantee that the submissions do not contain any such occurrences. As a result of artificial intelligence capabilities, AIVS can not only maintain track of past academic data, but it can also actively defend the academic evaluation process against evolving fraudulent actions. The AIVS proposed here aims to ensure the legitimacy of academic records by combining the decentralized immutability of blockchain technology with the proactive fraud detection capabilities of artificial intelligence. This will be accomplished using AIVS. The architecture is made up of four main components: a blockchain ledger built on the Ethereum platform, smart contracts for managing academic transactions, anomaly detection modules based on artificial intelligence, and a user-friendly web-based frontend supported by a RESTful backend API. All these components are necessary for the architecture. When combined, these components form a comprehensive ecosystem that allows for the secure issuance of academic credentials, as well as the verification and monitoring of such credentials in real time. The dynamic link between these modules ensures that each academic record submitted into the system is not only encrypted for security but also continuously monitored for any inconsistencies that may occur. Artificial intelligence modules evaluate patterns and identify potentially dishonest actions, ensuring a dynamic approach to sustaining academic honesty. While artificial intelligence modules analyze patterns and identify instances of potentially fraudulent conduct, blockchain provides a framework for securely storing acquired data.

A decentralized ledger supports academic accomplishments. To protect privacy, the system refrains from directly storing raw academic data. Each academic record is initially hashed using the SHA3-512 cryptographic method before its storage on the Ethereum blockchain. The hashing process $H_r$ is defined mathematically as follows:

$H_r=\text{SHA}3-512\left(A_{re}\right)H(1)$

where $H$ is the hash, $A_{re}$ represents the original academic record data. The hash ensures that even a single alteration in the academic records produces a vastly different H, which makes tampering immediately detectable. A transaction is created using a smart contract to insert a hashed record onto the blockchain. The transaction can be mathematically represented as:

$T_x=\left\{H,A_s,t\right\}(2)$

where $T_x$ is the blockchain transaction, $A_s$ is the sender’s blockchain address, and t is the timestamp at which the transaction was generated. The Ethereum network uses a consensus process to reach an agreement on the state of the ledger. During the development phase, Ganache is used to recreate a local blockchain environment, allowing for controlled and efficient testing of all contract capabilities. Smart contracts, which operate as autonomous entities, supervise the verification, creation, and administration of academic records. Solidity has meticulously developed all of the functions that comprise a smart contract. A smart contract is used to add a record to the blockchain, which can be described as follows:

$Insert\left(H\right)=require\left(msg.sender=University\right)\wedge store\left(H\right)(3)$

where $Insert\left(H\right)$ is the function to insert a record, $msg.sender$ is the address initiating the function call, and the University is the authorized address allowed to add records and store(H) saves the hash H into the blockchain storage. Record verification is performed when an external verifier wishes to confirm the authenticity of an academic record. This process is modeled as:

$Verify\left(H^{\prime }\right)=\left(H^{\prime }=H_s\right)(4)$

where $H^{\prime }$ is the newly generated hash, and $H_s$ denotes the stored hash for the blockchain. If H′ matches with $H_s$ , then the record is considered authentic. An audit trail is maintained to track all access to academic records. Thus, the audit log $A_l$ can be determined as follows:

$A_l=\left\{U_a,H_a,t_{st}\right\}(5)$

where $A_l$ denotes the audit log entry, $U_a$ is the record accessed by the user, $H_a$ is the associated record hash, and $t_{st}$ is the timestamp of access. Thus, the proposed approach ensures strict role-based access control, immutability of academic achievements, and full traceability of any interactions with stored credentials.

On the other hand, artificial intelligence modules are necessary for the active monitoring of academic patterns to identify possible fraudulent actions. Blockchain technology offers tamper resistance. The data on academic achievement are initially represented as a time series in the beginning of the anomaly detection model:

$S=\{g_1,g_2,g_3,\dots ,g_{n\}}(6)$

In this context, the letter $S$ represents the ordered sequence of grades that a specific student has earned overall across a variety of academic examinations.

At the $i^{th}$ evaluation point, a grade is attained, with each element $gi$ denoting that grade. This framework enables the simulation of temporal trends and variations in student performance over semesters, courses, or projects. The grade volatility $V$ is computed by the system to detect any fluctuations in academic performance that may warrant attention.

$V=\sqrt{\left\{\frac{1}{n-1}{\displaystyle \sum _{\left\{i=1\right\}}^n}{\left(g_i-\overline{g}\right)}^2\right\}}(7)$

where $\overline{g}$ is the mean of all grades and higher value of $V$ may indicate inconsistent performance deserving further investigation. Predictive modeling is utilized to anticipate expected grades based on historical data through linear regression.

${\overline{g}}_{i+1}={\beta }_0+{\beta }_1\times i(8)$

where g ̅_(i+1) denotes the predicted grade for the next academic event, ${\beta }_0+{\beta }_1$ are model coefficients learned during training i is the index of the current academic period. To evaluate the consistency of student performance and detect any suspicious deviations, a predictive model is employed to forecast expected grades. The residual error e is subsequently determined as:

$e=g_{i+1}-{\widehat{g}}_{i+1}(9)$

Equation 9 ascertains the discrepancy between the predicted and actual grades for the forthcoming academic event. A significant residual $e$ suggests that the student’s present performance diverges from historical trends. This may signify wrongdoing, manipulation, or grade inflation. If the residual e exceeds a specified threshold $\theta$ An anomaly alert is activated.

$Alert=\left\{\begin{array}{c}1,If\mid e\mid >\theta \\ 0,otherwise\end{array}\right.(10)$

In addition to numeric data, textual information such as thesis titles and project reports is analyzed using natural language processing. The semantic similarity between documents is assessed using cosine similarity

$CosSim\left(d_1,d_2\right)=\frac{d_1.d_2}{\parallel d_1\parallel \times \parallel d_2\parallel }(11)$

where $d_1.d_2$ are the document vectors; a low cosine similarity score between sequential submissions may indicate suspicious inconsistency in a student’s writing style. The combination of numeric volatility, prediction residuals, and semantic anomalies creates a comprehensive fraud detection framework that is continuously learning and adapting. Machine learning models, including isolation forests and support vector machines, are utilized to further refine anomaly classifications based on evolving data patterns. Finally, the anomaly detection model outputs a risk score $R_s$ for each student:

$R_s={\lambda }_{1}V+{\lambda }_2\left|e\right|+{\lambda }_3\left(1-CosSim\right)(12)$

where $R_s$ is the composite risk score, ${\lambda }_1,{\lambda }_2\lambda ,{\lambda }_3$ weights are assigned to each anomaly component based on its importance. A higher $R_s$ indicates a greater probability of fraudulent behavior requiring human review. This layered approach ensures a robust, data-driven method for safeguarding academic integrity while minimizing false positives. Therefore, the proposed AIVS combines blockchain’s immutability, smart contracts’ automation, and AI’s proactive detection mechanisms to create a secure and scalable solution for academic record verification.

Figure 2 depicts the architecture of the proposed academic integrity verification system, which combines blockchain technology, smart contracts, and AI-driven fraud detection to ensure the immutability and trustworthiness of academic data. At the top, the integration and verification cycle shows how academic components are constructed, tested, and certified at several tiers before being accepted into the verified system. The diagram depicts a blockchain component that uses SHA-512 data hashing to protect records from manipulation. Smart contracts automate record verification and access control, resulting in transparent, rule-based operations. AI-based anomaly detection servers discover data anomalies, while RNN and NLP models are used to detect semantic and temporal outliers, allowing for more robust fraud detection. Together, these components allow for an end-to-end, immutable academic record verification procedure that ensures data authenticity, traceability, and security.

Figure 2

Figure 2. Research methodology overview for AIVS implementation.

3.1 Justification of deployed algorithms

The choice of specific algorithms in this work was influenced by the dual imperatives of safeguarding academic records and ensuring the precision of anomaly detection. We chose each algorithm after comparing it to other, more regularly used ones to see how well it worked and how well it may be used in higher education. SHA3-512 was utilized for cryptographic hashing instead of more prevalent methods such as SHA-256 or MD5. SHA3-512 has an extended digest length and superior collision resistance, enhancing the likelihood of data integrity and detection of tampering inside the blockchain framework. This choice is especially important for managing academic records, where long-term security is quite important. The trade-off is a minor increase in computational overhead compared to SHA-256. This is acceptable because academic transaction data is relatively tiny, and immutability is a top goal in this field (Chaudhari and Shirole, 2025). Distance in Wasserstein. The Wasserstein Distance was used instead of Euclidean or cosine distance metrics for jobs that involved finding semantic similarities and anomalies. This metric’s strength is that it can find changes in the distribution of text embeddings, which makes it better at spotting subtle occurrences of copying, paraphrasing, or semantic drift. This feature is very important for judging academic writing, because little changes might hide problems. The trade-off, however, is that it is harder to compute, especially in large-scale analyses. In our implementation, nevertheless, batch optimization approaches helped with this (Chinnasamy et al., 2025). Dynamic Time Warping (DTW). DTW was used for sequence alignment tasks, including looking at how academic activities or submission practices change over time. DTW is different from normal similarity measures since it can align sequences of different durations and time distortions, which might help find underlying problems in submission timelines. The downside is that it takes more time to compute, which is a quadratic time complexity. However, this expense is worth it because it can find anomalies that linear approaches would miss (Balobaid et al., 2023). These algorithmic selections are consistent with recent survey results that underscore the necessity for resilient, context-sensitive methodologies in blockchain-based academic integrity systems. Research, including (Chaudhari and Shirole, 2025) and related studies on blockchain-based electronic educational document management (Chinnasamy et al., 2025) and blockchain with encryption for secure academic record preservation (Balobaid et al., 2023), underscores the necessity of choosing methods that harmonize accuracy, transparency, and scalability in educational systems.

4 Proposed academic integrity verification system AIVS

The academic integrity verification system is designed as a comprehensive solution for protecting academic records from manipulation and fraud. It combines blockchain technology, artificial intelligence, and advanced cybersecurity techniques into a cohesive architecture. The proposed AIVS is built on multiple technological layers, including an ASP.NET Core MVC and Bootstrap frontend, an ASP.NET Core 8.0 backend with RESTful APIs, a PostgreSQL database with Redis caching for efficient academic data storage, an Ethereum-based blockchain for record immutability, and TensorFlow with Keras frameworks for AI-driven anomaly detection. AIVS assures that once created, academic records are protected at all stages, from original entry to final verification. Each student’s academic history is securely recorded using cryptographic techniques that are extremely resistant to tampering or illegal changes. The system’s operations are based on advanced mathematical models and predictive algorithms that dynamically detect suspicious patterns and behaviors, increasing the legitimacy and transparency of academic achievements.

To ensure tamper-resistant academic data registration and efficient fraud detection, AIVS incorporates a mathematically rigorous model based on advanced cryptographic commitments, time series anomaly prediction, and semantic analysis of academic documents. At the core of blockchain-based data immutability, we construct commitment hashes $h_i$ for each academic record $D_i$ using an enhanced hashing function that incorporates timestamps:

$h_i=SHA3-512\left(D_i\parallel t_i\right)(13)$

where $t_i$ represents the exact timestamp of record creation. SHA3-512 ensures enhanced collision resistance, which improves the system’s resilience against future quantum threats. To reduce blockchain congestion and optimize record verification, AIVS aggregates individual academic records into a structured binary Merkle Tree. The total number of tree levels $L$ for $n$ records satisfy:

$L=\lceil {\mathit{log}}_2\left(n\right)\rceil (14)$

The equation $L=\lceil {\mathit{log}}_2\left(n\right)\rceil$ determines the total number of levels $L$ required to build a Merkle Tree from $n$ academic records. Here ${\mathit{log}}_2\left(n\right)$ calculates the base-2 logarithm of the total records, representing how many times the dataset must be divided by two to reach a single root hash. The ceiling function $\lceil ·\rceil$ rounds up to the nearest whole number because the number of levels must be an integer; partial levels are not possible in the tree structure. This ensures that during verification, every academic record can be reached efficiently, in logarithmic time complexity $\mathit{log}\left(n\right)$ thereby maintaining system scalability even as the number of records grows.

Furthermore, the probability of a successful record tampering attempt is inversely proportional to the cumulative proof-of-work difficulty $D$ in the underlying blockchain network, mathematically expressed as:

$P_{tamper}\approx \exp \left(-\frac{D}{K}\right)(15)$

Here $P_{tamper}$ represents the probability of a tampering attempt being successful, $D$ denotes the cumulative proof-of-work difficulty in the underlying blockchain network, and $K$ is the average computational power accessible to a potential adversary. This model demonstrates that as the blockchain network’s cumulative difficulty $D$ increases relative to an adversary’s computational capacity $K$ , the probability of successful tampering diminishes exponentially. Hence, securing academic records on a blockchain with higher mining difficulty provides significantly greater resistance against unauthorized modifications. For academic anomaly detection, the AIVS system models each student’s grade trajectory over time using a multivariate Gaussian process:

$g\left(t\right)\sim GP\left(\mu \left(t\right),K\left(t,t^{\prime }\right)\right)(16)$

Here $GP$ stands for Gaussian Process, $g\left(t\right)$ denotes the predicted grade at time $t,\mu \left(t\right)$ is the mean function that represents the expected academic performance trend at a given time, and $K\left(t,t^{\prime }\right)$ is the covariance kernel that quantifies the similarity or correlation between student performances at different time points $t$ and $t^{\prime }$ . By capturing both the trend and variability of academic performance, this probabilistic model enables the early identification of atypical academic behavior. Deviation from the expected academic performance is further measured using the Mahalanobis distance:

$d_M\left(x\right)=\sqrt{{\left(x-\mu \right)}^T}\sum -1\left(x-\mu \right)(17)$

Here $d_M\left(x\right)$ represents the Mahalanobis distance of an observed academic performance vector $x$ from the expected mean performance $\mu$ $\sum$ is the covariance matrix characterizing the variability and correlation structure among academic records, and $T$ indicates the matrix transpose operation. A high Mahalanobis distance indicates that a student’s academic record significantly deviates from the established norm, potentially signaling suspicious patterns such as grade inflation, academic misconduct, or unauthorized interventions. By applying this statistical distance metric, AIVS ensures a robust and context-sensitive evaluation of student performance anomalies. Figure 3 depicts the temporal trend of academic activity and anomaly scores across semester-specific academic activities in the AIVS system. The solid blue curve depicts the projected grade progression of a student over time. The orange dashed line represents the AI-generated risk score, which varies according to observed abnormalities. Triangular markings along this line signify threshold breaches, which occur when the algorithm detects large departures from predicted patterns and issues an alarm. The black circles represent the timestamps at which academic records are committed to the blockchain, which ensures data immutability. Purple X-marks indicate areas where administrative involvement is required, indicating probable academic integrity issues. Furthermore, the symbol (S) represents instances of semantic drift detection utilizing NLP algorithms, which capture changes in textual or contextual academic data. The figure shows how AI and blockchain work together to monitor academic conduct and identify abnormalities for evaluation.

Figure 3

Figure 3. Shows how the AI module in AIVS monitors academic behavior over time, detecting anomalies, semantic drift, and risk patterns for proactive misconduct prevention.

To model the semantic consistency of students’ written submissions over time, AIVS generates sentence embedding’s and evaluates the semantic distance between successive documents using the Wasserstein Distance, mathematically defined as:

$W\left(p,q\right)=\gamma \in \mathrm{\Gamma }\left(p,q\right)\mathit{inf}{E}_{\left(x,y\right)}\sim \gamma \left[\parallel x-y\parallel \right](18)$

where $p$ and $q$ denotes the probability distributions of sentence embedding are from two consecutive academic submissions and $\mathrm{\Gamma }\left(p,q\right)$ denotes the set of all possible joint distributions whose marginals are $p$ and $q$ and $\gamma$ a specific coupling (mapping) from $p$ to $q$ that minimizes the expected cost, $\left(x,y\right)$ are sampled embedding pairs, and $\parallel x-y\parallel$ is the Euclidean distance between them and $\mathit{inf}$ indicates the infimum (greatest lower bound) over all possible couplings $\gamma .$ Thus, the Wasserstein Distance measures how much “work” is needed to morph one document’s semantic distribution into another, offering a robust method to detect major stylistic inconsistencies suggestive of plagiarism or ghostwriting. For time-based performance consistency, AIVS applies the dynamic time warping DTW distance between academic grade sequences:

$DTW\left(S,T\right)=\mathit{min}\sqrt{{\displaystyle \sum _{\left(i,j\right)\in \pi }}{\left(S_i{T}_j\right)}^2}(19)$

where $DTW$ denotes the dynamic time warping, $S$ and $T$ are sequences of academic performance scores (e.g., grades) over time, $\pi$ represents the optimal warping path, which aligns elements of $S$ and $T$ to minimize cumulative distance and $S_i{T}_j$ are aligned pairs from the two sequences. The dynamic time warping allows AIVS to align and compare academic progressions even if students proceed through assessments at different paces, capturing abnormal variations.

An authentication session within AIVS uses the elliptic curve cryptography (ECC) for digital signatures. For a private key $k$ and public generator point $G$ , the signature $\left(r,s\right)$ on message $m$ is computed as:

$r={\left(kG\right)}_X{\mathit{mod}}^n,s=k^1\left(H\left(m\right)+rd\right)(20)$

Here $k$ is a random nonce chosen per signature, ${\left(kG\right)}_X$ is the x-coordinate of the elliptic curve point after scalar multiplication and $n$ is the order of the elliptic curve and $H\left(m\right)$ is the cryptographic hash of the message $m$ and the ${\mathit{mod}}^n$ means modulus operation with respect to $n$ which reduces the number to within the range $0ton-1.$ The verification requires checking represented as:

$\left(s^{-1}H\left(m\right)\right)G+\left(s^{-1}r\right)Q=\left(kG\right)(21)$

Here $s^{-1}$ is the modular inverse of the signature component $s$ under modulo $n$ . It is used to reverse the multiplication done during signature generation, $H\left(m\right)$ represents the hash of the original message mmm (e.g., the academic record or credential data). A secure hash function SHA3-512 is applied, $G$ is the generator point of the elliptic curve, which is publicly known and defines the curve’s structure and $r$ is the first part of the signature, derived from the x-coordinate of $kG$ during signature creation and $Q$ is the public key corresponding to the signer’s private key $d$ . It is used by verifiers to validate the signature without knowing the private key and the $kG$ is the elliptic curve point generated during signing using the random number $k.$ The left-hand side reconstructs the point $kG$ using only public information $\left(r,s,Q,G\right)$ and the hashed message $H\left(m\right)$ . If the reconstructed point matches the originally signed point, the signature is considered valid. Additionally, anomaly scores for academic behavior are aggregated using a weighted softmax function:

$P_i=\frac{e^{{\alpha }_i{x}_i}}{{\sum }_{j=1}^n{e}^{{\alpha }_j{x}_j}}(22)$

where $x_i$ denotes the normalized feature score for the $i^{th}$ indicator of academic behavior, such as grade deviation, writing style consistency, or submission timing, $\alpha$ is the importance weight assigned to feature iii. It reflects how critical that feature is when evaluating academic integrity and $e$ is the base of the natural logarithm, approximately 2.718, used for exponential scaling to magnify the influence of higher-scoring anomalies and ${\sum }_{j=1}^n{e}^{{\alpha }_j{x}_j}$ is the sum of exponentials over all features, ensuring that the output probabilities $P_i$ are normalized and add up to 1. It applies a softmax function to transform the raw, weighted feature scores into a probabilistic distribution that emphasizes stronger anomalies while still considering all factors proportionally. The final risk score $R$ for each student is computed as a weighted sum:

$R={\displaystyle \sum _{i=1}^n}{w}_i{P}_i\mathrm{R}>\mathrm{\tau }(23)$

Here $w_i$ is the weight assigned to each softmax probability $P_i$ . These weights reflect institutional priorities, for example, some institutions may consider grade inflation more serious than minor submission delays, $P_i$ is the probabilistic anomaly score for feature iii from the softmax output and $\mathrm{R}$ is the composite risk score for a student’s academic behavior, representing the overall risk level and $\mathrm{\tau }$ is the risk threshold a tunable parameter set based on historical analysis. If $\mathrm{R}$ exceeds $\tau$ , the academic profile is flagged for further administrative review. This step aggregates the weighted anomaly probabilities to compute a single risk score $R$ . If the risk exceeds a predefined threshold $\tau$ It indicates potential academic misconduct requiring human investigation, $w_i$ reflects institutional priorities (e.g., more weight on grade irregularities versus writing style anomalies) and an academic credential is flagged for administrative investigation if $\mathrm{R}>\mathrm{\tau }$ is a tunable risk threshold set based on historical analysis.

AIVS uses elliptic curve digital signatures to verify academic records by reconstructing the original signing point using modular inverses, hashed academic data, and public key information. A weighted softmax function normalizes anomaly indications into probabilistic scores to quantify academic behaviors after verification, highlighting institutionally important aspects. Softmax-transformed scores are pooled through a weighted summation model to calculate each academic profile’s composite risk score. Academic records are automatically identified for administrative investigation if computed risk exceeds a threshold. Cryptographic verification, anomaly probability normalization, and risk scoring form a comprehensive framework for proactive, real-time academic irregularity identification while retaining the highest data security and operational efficiency.

Algorithm 1

Algorithm 1. Secure academic record commitment.

The commitment phase commences with Algorithm 1. The incorporation of academic records into blockchain results in this outcome. This occurs before the submission of the academic record. One of its responsibilities is the organization of educational records. The method initiates by integrating the academic transcript with its corresponding timestamp. This constitutes the initial stage. This generates a distinctive message, referred to as MMM, at the conclusion. In the second stage, the SHA3-512 algorithm cryptographically hashes message M to generate an immutable commitment hash. Modifications are prohibited. The recipient is then provided with this hash.

The hash represents the ultimate result of the operation. Figure 4 illustrates the layered architecture of the proposed academic integrity verification system, showing how technology components work together to safeguard and intelligently handle academic records. Top-level user interfaces allow students, professors, and administrators to engage with the system. The AI module is linked to the application layer’s ASP.NET Core MVC frontend and RESTful API backend. The TensorFlow and Keras-based AI module detects anomalies and provides academic risk levels. Data management employs PostgreSQL to store academic data and Redis for caching to improve performance under the application layer. Ethereum smart contracts and SHA3-512 cryptographic hashing enable academic record immutability and verifiability at the blockchain layer. These components create a strong, scalable, and secure anomaly detection and academic integrity system.

Figure 4

Figure 4. Architecture of AIVS: An integrated system of user interfaces, application components, data storage, AI modules, and blockchain layers for academic integrity verification.

Algorithm 2 illustrates the computation of the academic risk score.

Algorithm 2

Algorithm 2. Computation of the academic risk score.

Algorithm 2 calculates a risk score from academic data to predict academic underperformance or worry. A matrix of academic features, relevance levels for each item, weights, a variable for the final risk score, and a threshold value are initialized. To standardize data, the feature matrix is normalized. Each feature’s contribution is calculated using its normalized value and relevance level, with an exponential-based method that highlights important features. The total influence of all features is calculated from these contributions. Summing each feature’s weight and contribution value yields the risk score. Finally, the threshold is compared to this score. If the risk score exceeds the threshold, it may indicate an academic difficulty.

The implementation of the artificial intelligence verification system ensures that academic record verification is conducted in a comprehensive and scalable manner. The c-driven anomaly scoring Algorithm 2 and the cryptographic commitments Algorithm 1 are coupled and integrated, and then rigorously merged and integrated into each other to achieve this goal. The enhanced confidence that can be found throughout the entire academic ecosystem is a direct result of this dual-layered method, which provides a strong defense against both unauthorized record tampering and developing trends in academic fraud. Overall, this technique is responsible for the increased confidence that can be found. Figure 5 shows the blockchain-AI academic record verification workflow. The main user interfaces for students, instructors, and administrators submit academic records using a RESTful backend API. SHA3-512 is used to build a cryptographic hash from submitted grades. This hash is submitted to an Ethereum-based blockchain network via smart contracts to keep records immutably and tamper-proof. Additionally, an academic database stores the original academic records. An AI-based anomaly detection module checks the stored hash for inconsistencies and unusual activity. Alarms and risk assessments are shown on an administrative dashboard from detection findings. The system cross-validates stored data and blockchain hash, checks for AI-generated alarms, and reacts to record verification requests to ensure transparency, integrity, and early academic fraud detection.

Figure 5

Figure 5. AI-blockchain workflow for academic record verification.

The TensorFlow and Keras-based AI anomaly detection module uses the academic database and blockchain ledger. Real-time analysis of student academic trends reveals grade abnormalities, semantic discrepancies in writing, and time-based submission irregularities. Grade inflation, ghostwriting, and academic fraud are detected using multivariate time series analysis and semantic drift detection. The administrative risk dashboard aggregates and visualizes AI analysis results, assigning comprehensive risk scores to academic profiles and automatically flagging abnormalities. Thus, administrative authorities can make prompt, evidence-based interventions without compromising fairness or openness. The proposed AIVS creates a strong, scalable, and proactive academic integrity system by integrating user engagement, blockchain immutability, AI-driven anomaly detection, and administrative oversight. It protects academic records from manipulation, dynamically detects fraud, and boosts trust between academic institutions and external verifiers.

4.1 Consolidated hybrid AI–blockchain architecture

It integrates the conclusive description of the proposed hybrid architecture into a cohesive structure. The academic Integrity Verification System amalgamates blockchain technology, smart contracts, and artificial intelligence modules into a cohesive framework that guarantees immutability and sophisticated anomaly detection. This offers a detailed examination of its structure, training methodology, operational procedures, and mathematical framework.

⁃ Blockchain Ledger

The blockchain ledger provides secure and immutable storage of academic records. Every record is hashed via SHA3-512, integrated with a timestamp, and recorded on the Ethereum blockchain. The system utilizes a Merkle tree structure to ensure scalability, facilitating the efficient verification of extensive academic material. The unchangeability of blockchain transactions ensures that once a record is appended, it cannot be modified without detection, thereby preserving academic integrity at the data storage level. The hash $H$ and parent hash $H_p$ can be calculated as follows:

$H=SHA3\left(512\right)\left(R\parallel t\right);H_p=SHA3\left(512\right)\left(L_{left}\parallel L_{right}\right)(24)$

where $R$ denotes the record, and $t$ is the timestamp. The verification uses Merkle root.

⁃ Smart contracts

It is authored in Solidity and executed on Ethereum, regulating all essential functions. They verify the legitimacy of supplied records by examining student identities, course codes, and submission timestamps. Additionally, they implement role-based permissions, guaranteeing that only authorized entities can edit or modify records. Every interaction is recorded as an immutable audit trail, enhancing accountability and transparency. Smart contracts initiate anomaly detection protocols during verification requests, integrating blockchain security with artificial intelligence.

The recomputed hash $H_{bl}$ value and the hash value stored on the blockchain $H_{bl}$ are computed as follows:

$H^{\prime }=SHA3\left(512\right)\left(R\parallel t\right);H^{\prime }=H_{bl}(25)$

⁃ Artificial Intelligence Layer

The artificial intelligence layer enhances the unchangeable blockchain storage by enabling dynamic and proactive fraud detection through the constant analysis of numerical and textual academic data. This layer integrates three complementary approaches to guarantee thorough coverage of probable abnormalities. Recurrent Neural Networks (RNNs) are utilized to analyze student grade sequences over several semesters, enabling the system to identify temporal relationships in performance and detect sudden anomalies that may indicate manipulation or irregularities. Natural language processing (NLP) models analyze textual submissions, including theses, reports, and project documentation, by evaluating writing style, semantic coherence, and content similarity. These models assist in detecting instances of plagiarism, ghostwriting, or abrupt semantic shifts that signify a deviation from a student’s established writing style. The results from RNN-based grade analysis and NLP-based text evaluation are subsequently integrated within an Anomaly Scoring Engine. This engine consolidates indicators, including grade volatility, prediction residuals, and semantic inconsistencies, into a singular composite score through weighted functions. The final risk score assesses the probability of fraudulent activity, allowing administrators to concentrate on cases that surpass established thresholds while ensuring efficiency and impartiality in the verification process.

The predicted grade for the student $g^{\forall }+1$ , the prediction error $e_i$ +1 can be determined as follows:

$g^{\forall }+1=f_{\theta }\left(\mathrm{g}:\mathrm{i}\right);e_i+1=g_i+1-g^{\forall }+1(26)$

The vector representation of a document, denoted as $v$ , and the comparison with another document, $w$ , can be established as follows:

$\mathit{cos}\left(v,w\right)=\frac{v.w}{‖v‖‖w‖}(27)$

Flagging anomalies when $\mathit{cos}\left(v,w\right)<\gamma$ where $\gamma$ is a predefined threshold value

⁃ Integration and Training Process

Synthetic datasets that reflect anonymized student profiles, grades, and thesis submissions are used to integrate the components. The AI modules utilize TensorFlow and Keras frameworks for training, whereas blockchain smart contracts serve as anchors for record authenticity. During execution, smart contracts invoke AI routines upon verification requests, and AI alarms are subsequently recorded on the blockchain to guarantee traceability. This approach enables the system to perpetually adapt, enhancing fraud detection precision as additional data is acquired.

Let $p_j$ be the normalized SoftMax weight that shows the relative significance of the feature $j$ , the scaling parameter $a_j$ is applied to each feature $z_j$ . Weight is assigned to each feature, and the total number of features $t{z}_j$ should be considered. Thus, the final composite risk $R_c$ can be calculated as follows:

$p_j=\frac{\mathit{exp}\left(a_j,z_j\right)}{{\displaystyle {\sum }_{k=1}^{t{z}_j}}\mathit{exp}\left(a_k,z_k\right)};R_c={\displaystyle \sum _{j=1}^{t{z}_j}}{w}_j{p}_j(28)$

where $k$ denotes the index of summation in the denominator.

Let $T_{ob}$ represent the training objective, and $C_l$ denote the classification loss utilized for anomaly detection tasks. The grade prediction task, denoted as ${\mathit{Pr}}_g$ , is equilibrated ψ through the implementation of classification and regression tasks. The regularization coefficient $\tau$ which is necessary for regulating the penalty term. Weight regularization can be achieved by using a regularization coefficient. The model parameters necessitate a compromise between regularization to avert overfitting.

$T_{ob}=\mathrm{\psi }{.C}_l+\left(1-\mathrm{\psi }\right){\mathit{Pr}}_g+\tau {‖\theta ‖}_2^2(29)$

⁃ End-to-End Workflow

The unified procedure begins with the creation and submission of an academic record. The record is hashed with the SHA3-512 function and a timestamp, then permanently inscribed on the blockchain to guarantee immutability and data security. Upon record placement, a smart contract promptly recalculates the hash and compares it with the blockchain entry to ensure submission integrity. Smart contract authenticates student identities and enforces role-based restrictions to guarantee that only authorized users can submit or modify information. Upon validation, AI modules analyze the record. Recurrent neural networks evaluate performance trends in numerical data, such as semester grades, to identify abrupt deviations from anticipated trajectories. Natural language processing algorithms examine theses, reports, and assignments for instances of plagiarism, ghostwriting, and substantial stylistic alterations. The anomalous scoring engine employs weighted SoftMax aggregation to integrate grade volatility, prediction errors, and semantic drift into a composite risk score derived from numerical and textual data. A risk score estimates the likelihood of anomalous or fraudulent behavior. An administrative dashboard featuring blockchain-supported audit records and AI-generated alerts presents the outcomes. This interface enables administrators and academic authorities to monitor activities in real time, identify the sources of anomalies, and make evidence-based decisions to uphold academic integrity.

5 Experimental evaluation and results

This section provides the experimental evaluation and results of the proposed academic integrity verification system.

5.1 Test environment and replicability

The academic integrity verification system was instituted in a regulated simulation environment to facilitate reproducibility. Synthetic datasets, including 500–1,200 anonymized student profiles, were utilized to simulate academic records, simulating enrollment histories, grade distributions, and thesis submissions. The experimental environment comprised an Ethereum blockchain layer (smart contracts deployed through the Ganache simulator and the Ropsten Proof-of-Authority test network for regulated transaction execution) and an AI layer constructed using TensorFlow 2.0 and Keras frameworks, which trained recurrent neural networks (RNNs), regression models, natural language processing (NLP) modules, and anomaly detection algorithms. The configuration comprised an Intel Core i7 processor, 32 GB of RAM, an NVIDIA RTX 3080 GPU, and Python 3.10 operating on Ubuntu 22.04. Assessment criteria centered on verification duration, anomaly detection precision, system resource consumption, user contentment, and blockchain storage burden.

This precise specification enables the replication of results in an analogous computing environment. A shortcoming of the existing model is the lack of integration between course-level and instructor-level grading histories, potentially constraining its predictive capability. Subsequent research will enhance the framework by integrating these contextual aspects to yield more reliable predictions. Table 2 illustrates the specification of the experimental setting and replicability.

Table 2

Table 2. Experimental environment and replicability specification.

5.2 Experimental evaluation and results

The experiment was carried out, and the findings showed that the suggested AIVS system performed exceptionally well across all the critical factors. Academic data integrity was effectively kept, fraud was detected with high precision, verification time was greatly decreased, user satisfaction was increased, and adequate resource management was provided during peak demand periods. All findings supported AIVS’s scalability, durability, and practical application in higher education settings. The following important indicators were judged crucial following the evaluation of the experiment’s outcomes:

⁃ Blockchain record integrity verification

⁃ AI-based anomaly detection performance

⁃ Verification time efficiency analysis

⁃ User satisfaction across stakeholder groups

⁃ System resource utilization and scalability

⁃ Anomaly Detection Accuracy

⁃ Performance Comparison of AIVS Model with AI and Existing Plagiarism Detection Systems

⁃ Storage Overhead and Frequency

⁃ Blockchain Storage and Cost Analysis

5.2.1 Blockchain record integrity verification

A blockchain record integrity evaluation was performed to confirm the proposed AIVS system’s tamper-resistance capabilities. Each academic record was cryptographically encoded and then saved on the Ethereum blockchain using smart contracts. Subsequent integrity checks on 500 student records demonstrated a 100% success rate in ensuring the immutability of stored records. No unauthorized adjustments were found after recording, showing the efficiency of blockchain immutability in safeguarding academic data integrity. Figure 6 shows that all 500 academic records added to the blockchain were validated as tamper-free over a 6-month simulated period. The lack of detectable changes validates the immutability guarantee provided by blockchain integration, assuring that once academic material is recorded, it is forever secure and unalterable. This study lends credence to the usage of blockchain as a secure foundation for storing academic credentials.

Figure 6

Figure 6. Blockchain record integrity verification results.

5.2.2 AI-based anomaly detection performance

The performance of the AI anomaly identification module was assessed by introducing controlled anomalies into the academic dataset, such as rapid grade changes and inconsistent academic behavior. Figure 7 shows that the system has a high detection rate of 94% for fraudulent entries, with an overall model accuracy of 95%.

Figure 7

Figure 7. Performance metrics of the proposed AIVS anomaly detection module, showing a 94% detection rate for fraudulent entries and an overall model accuracy of 95%.

These results demonstrate the AI engine’s robustness in distinguishing unusual academic practices. The findings demonstrate a high true positive rate, indicating that most actual fraud cases were successfully discovered. The result indicates that regular student performance was rarely misidentified. This demonstrates that the AI model within AIVS is extremely reliable, both in detecting fraudulent conduct and minimizing wasteful investigations.

5.2.3 Verification time efficiency analysis

To statistically assess the operational efficiency of the blockchain, performance parameters like gas consumption, block confirmation time, and transaction throughput were documented during the Ganache and Ropsten test tests.

A key performance indicator was comparing traditional academic record verification methods with the AIVS blockchain-based approach. Traditional manual verification, which involved paperwork and communication between institutions, was significantly less efficient. AIVS cut verification times by 85%, confirming transactions in about 5 s during the Ganache simulation environment. This impressive improvement highlights the efficiency and practical application of the system.

Figure 8 clearly shows the major decrease in verification times through the AIVS platform. Manual verification, which could take minutes or hours depending on institutional responses, was replaced by almost immediate blockchain validation. This performance boost demonstrates AIVS’s potential to simplify academic credential verification and greatly enhance operational efficiency in academic administration.

Figure 8

Figure 8. Academic record verification time comparison.

The efficacy of the Ethereum-based blockchain layer was additionally assessed utilizing both the Ganache simulator and the Ropsten Proof-of-Authority (PoA) test network. In these studies, the average block confirmation time on Ropsten was roughly 14 s, with a mean gas usage of approximately 85,000 units per transaction and an average transaction cost of about 0.011 USD during periods of low congestion. The Ganache local testnet attained a transaction throughput of 23 transactions per second (TPS) without significant latency. When scaled to 1,200 simulated academic records, the blockchain exhibited consistent speed, verifying all data in an average of under 6 s.

The analysis focused on the integration of Layer-2 rollup frameworks, such as Optimistic Rollups on Arbitrum and zk-Rollups on Polygon, to improve scalability and cost effectiveness. Initial assessments suggest that transferring high-frequency activities, such as record verification and audit logging, to Layer-2 may decrease gas expenses by 60%–90% and enhance throughput to over 100 TPS, while maintaining the immutability and auditability of the Ethereum mainnet for credential commitments. This hybrid design guarantees that AIVS remains fiscally sustainable and computationally efficient for nationwide implementation across various colleges.

5.2.4 User satisfaction across stakeholder groups

Structured questionnaires sent to students, instructors, and administrators were used to assess end-user satisfaction. AIVS has an excellent 92% overall acceptance rating. Students were particularly impressed by the transparency and quickness of the verification process, while instructors and administrative professionals acknowledged the system’s potential to minimize administrative responsibilities despite small initial learning challenges. These data illustrate the system’s usefulness and widespread popularity among educational stakeholders. Figure 9 presents the user satisfaction survey results for each important stakeholder group. Students expressed the greatest pleasure, citing the platform’s ease of use and transparency. Instructors and administrators reported similar, albeit slightly lower, results, owing mostly to the early adjustment to blockchain-based operations. Overall, the system’s high approval ratings reflect its accessibility, dependability, and usefulness in a university setting.

Figure 9

Figure 9. User satisfaction levels by role.

5.2.5 System resource utilization and scalability

The efficiency of AIVS was also measured in terms of resource consumption under concurrent load circumstances. Even with up to 100 active users, backend server CPU utilization stayed below 45%, while AI module memory consumption hovered around 2 GB throughout live anomaly detection procedures. The blockchain storage overhead was low, confirming the lightweight approach of storing only cryptographic hashes rather than complete documents.

These results show that the AIVS design is scalable for real-world implementation. Figure 10 depicts the breakdown of system resource use during simulated moderate-to-high user demands. The results show that server resources are managed efficiently, with CPU and memory consumption remaining within ideal operational boundaries and blockchain storage overhead being minimal. This indicates that the AIVS platform can scale to bigger academic environments without requiring costly infrastructure improvements.

Figure 10

Figure 10. System resource utilization under load.

5.2.6 Anomaly detection accuracy

Figure 11 depicts four types of AI anomaly detection outcomes within the AIVS architecture. The first bar, labeled True Positives, is substantially higher than the rest, showing 92%. True Negatives and False Positives are both small, at 3%, indicating that the model is balanced in its correct and incorrect judgments of genuine behavior. The final indicator, False Negatives, is 2%, indicating that no cases of academic fraud were overlooked by the AI engine. The bars are colored orange and marked with percentages above each column to show the distribution accuracy of the AI model’s classification capability.

Figure 11

Figure 11. Distribution of anomaly detection outcomes in the AIVS model, demonstrating a high true positive rate (92%) and low false detection rates, validating the AI module’s accuracy and reliability in identifying academic irregularities.

5.2.7 Performance comparison of AIVS model with AI and existing plagiarism detection systems

The proposed AIVS is evaluated against RNN, NLP, and SVM classifiers using conventional classification metrics, including precision, recall, F1-score, confusion matrix, and ROC–AUC curves. The dataset was divided into 80% training and 20% testing sets using stratified random sampling to maintain a balanced class distribution. To prevent overfitting, each model was trained for 50 epochs with an early halting mechanism that utilized validation loss. The RNN and NLP models were enhanced in their ability to generalize by the addition of L2 regularization and dropout layers (0.3 probability). Tenfold cross-validation was implemented for the SVM classifier.

The evaluation dataset was designed to show how plagiarism and AI-assisted writing detection work in the real world. It was made up of three types of text: (i) real student theses written without AI help, (ii) AI-generated documents made with large language models, and (iii) adversarial altered samples where AI-generated content was manually changed to look like human writing styles and avoid detection. The dataset was balanced to include both native and non-native English writing styles, which is what you would expect to see in Kazakhstani higher education. We tested Turnitin, GPTZero, Copyleaks, and the suggested model on the same dataset and in the same way. We used each platform’s built-in interface or API to parse the input texts and record the detection results. Then, these outputs were turned into binary categorization outcomes (AI-generated vs. human-authored) so that all the tools would be consistent. Following typical information retrieval formulas, we calculated performance indicators including accuracy, precision, recall, F1-score, false positive rate, and false negative rate. We purposely added situations involving paraphrase, synonym replacement, and code-switching between English and Kazakh to make the surroundings more hostile. These cases examined the resilience of each system to nuanced manipulation and linguistic diversity. Figure 12 shows how the proposed AIVS model compares Turnitin, GPTZero, and Copyleaks on six evaluation metrics: accuracy, precision, recall, F1-score, false positives, and false negatives. The proposed AIVS model consistently surpasses current systems in accuracy (95%), precision (94%), recall (94%), and F1-score (94%), while simultaneously attaining the lowest false positive (3.0%) and false negative (2.0%) rates.

Figure 12

Figure 12. Performance comparison of the proposed AIVS model with Turnitin, GPTZero, and Copyleaks across accuracy, precision, recall, F1-score, false positives, and false negatives, showing superior results for the proposed model.

Table 3 shows the results, which show the pros and cons of the suggested model compared to systems that are frequently used. This makes sure that the evaluation is both methodologically sound and useful in real life.

Table 3

Table 3. Performance comparison of the proposed model with existing plagiarism detection systems in terms of accuracy, precision, recall, F1-score, false positives, and false negatives.

The ROC–AUC comparison of AIVS, RNN, NLP, and SVM trained for 50 epochs is depicted in Figure 13. The image illustrates that the AIVS model has the maximum ROC–AUC value of 0.97, which suggests that it is highly effective in distinguishing between classes and reducing false classifications. The RNN model approximates AIVS with a ROC–AUC of 0.96 and excellent generalization, but it has inferior sensitivity and specificity. A ROC–AUC of 0.95 is indicative of a stable but slightly diminished discriminative capability in an NLP-based model. The SVM model’s linear decision boundaries and incapacity to adapt to complex feature spaces are the reasons for its poor performance in classification, as evidenced by a ROC–AUC of 0.93.

Figure 13

Figure 13. Comparison of ROC–AUC values for AIVS, SVM, RNN, and NLP models, each trained for 50 epochs.

Figure 13 demonstrates that AIVS surpasses all other models in terms of prediction accuracy and robustness, thereby demonstrating its suitability for classification and decision-making in the experimental setup.

A distinct hierarchy in predictive effectiveness is evident in the comparison of model performance across key evaluation metrics (accuracy, precision, recall, and F1-score) as illustrated in Figure 14. The AIVS model consistently exhibits the highest overall performance, attaining a 95% accuracy, 94% precision, 94% recall, and 94% F1-score. These values suggest that AIVS not only predicts with high accuracy but also maintains a balanced trade-off between identifying true positives and averting false positives. The model’s robust reliability across a variety of testing conditions is indicated by its strong precision and recall, which implies that it is both accurate and stable in data classification. The RNN model exhibits competitive results, with a 93.4% accuracy, 93% precision, 93% recall, and 93.5% F1-score, which is marginally lower than AIVS. This slight decrease may be attributed to the sequential learning model’s tendency to occasionally result in minor overfitting or delayed convergence when contrasted with AIVS’s optimized architecture.

Figure 14

Figure 14. Comparison of performance metrics (Accuracy, Precision, Recall, and F1-Score) across five models: (AIVS, RNN, NLP, SVM, and AVIS). AIVS demonstrates the highest and most consistent results across all four metrics.

The NLP model exhibits consistent yet moderate performance, with a 93% accuracy, 92% precision, 91% recall, and 91.5% F1-score. Although it is possible for NLP methods to effectively capture semantic relationships, their efficacy may be limited by the complexity of feature dependencies in structured datasets. In contrast, SVM and AVIS demonstrate the lowest overall performance, with an accuracy of 90.5%, a precision of 90%, a recall of 89%, and an F1-score of 89.5%. These findings indicate that linear or static classifiers, such as SVM and AVIS, may not possess the adaptability and deep extraction capabilities required for dynamic classification tasks.

The proposed AIVS model outperforms both traditional and neural approaches in terms of classification efficiency across all evaluated metrics. Its consistently high scores serve as confirmation of its potential as a high-performing, generalizable, and reliable model for complex data-driven decision systems.

5.2.8 Storage overhead and frequency

Figure 15 shows the distribution of blockchain storage overhead in kilobytes over a collection of academic documents. The horizontal axis represents storage overhead in kilobytes, and the vertical axis depicts the frequency of records falling inside each range. Most entries, three in total, are between 0 and 5 KB in size, showing that most academic transactions maintained on the blockchain are lightweight and efficient, with merely hash values or minimum metadata.

Figure 15

Figure 15. Blockchain storage overhead across academic record sizes.

A few records require additional storage, with one occurrence each in the 5–10 KB, 10–15 KB, and 30–35 KB ranges. Notably, no records fall within the 15–30 KB interval, indicating the absence of medium-overhead entries. The pattern indicates an efficient storage strategy with occasional outliers, which could be attributed to extended metadata, contract complexity, or record aggregation circumstances. This improves the system’s scalability while revealing possible opportunities for optimizing larger transactions.

5.2.9 Blockchain storage and cost analysis

Figure 16 illustrates the correlation between record size and two significant measures. The measurements include blockchain storage overhead (KB) and projected gas cost (USD). The overhead expenditures related to storage and gas expenses exhibit a linear development trend when the record size escalates from 1 KB to 32 KB. The trends indicate a robust correlation between the volume of stored data and the associated computing expense. This indicates that larger record sizes result in proportionately higher storage demands and transaction costs. This link underscores the influence of storage overhead on the economic efficiency of blockchain-based solutions.

Figure 16

Figure 16. Comparative analysis of Blockchain storage overhead and estimated gas cost with increasing record size.

5.3 Simulation-based evaluation

To evaluate the feasibility of the planned academic integrity verification system, we created a controlled simulation in partnership with the International Information Technology University (IITU) in Almaty, Kazakhstan. Instead of using genuine student records, a synthetic dataset of 1,200 simulated student profiles was made to mimic the structure, variability, and activity patterns of real academic data. This dataset was made to seem like enrollment histories, grade distributions, thesis submissions, and verification requests, making it realistic without using personal or institutional details. A mock academic record management infrastructure was connected to the simulation environment through a RESTful API. This allowed AIVS to import and process fake records. To simulate stable production, records were hashed using SHA3-512 and stored on the Ethereum Ropsten test network, which uses Proof-of-Authority (PoA) consensus. An AI-driven anomaly detection module kept an eye on synthetic grades, submissions, and academic activity in real time. It found unusual trends and made sure that confirmed records could not be changed.

During the 2-month simulated operational period, 356 synthetic verification requests (from employers and academic institutions) were handled with an average delay of 4.8 s, which included checks using blockchain and AI. The test network figured out that each record would cost $0.011 to process, and the PoA consensus kept the costs very low. The simulated system was up 99.3% of the time, with no downtime longer than 10 min. The most concurrent usage was 87 verification or submission procedures without any drop in performance. Resource monitoring showed that the backend CPU use was below 46% even when the simulated demand was at its highest. The blockchain storage overhead was also very minimal because it only stored cryptographic hashes instead of the entire data.

There were 18 suspicious scenarios created in the synthetic anomaly detection evaluation, such as fake grade inflation and thesis contradictions. When these synthetic cases were looked over by hand, 83.3% were found to be true positives, 16.7% were found to be false positives, and none were false negatives. These results confirm that the system can uncover problems in a real-world setting. Simulated user feedback also showed that students liked being able to verify their own work, administrators liked having 70% less manual work to do, and employers liked verification that could not be changed. The accuracy of fraud detection was always over 94%, and the time it took to verify was always between 2 and 6 s in simulated network settings. Figure 17 depicts the verification latency distribution for the simulation of AIVS.

Figure 17

Figure 17. Verification latency distribution for simulation of AIVS.

6 Discussion of results

The experimental data confirms the proposed AIVS’s robustness, efficiency, and applicability. The tamper-resistance of academic records gained by blockchain integration shows that the system effectively mitigates one of the most serious risks to academic credibility: unauthorized data alteration. Achieving a 0% tampering rate over 500 records in 6 months demonstrates the immutability and dependability of Ethereum-based smart contract implementations for safe academic credential administration. The AI-based anomaly detection models contributed equally to the system’s overall success. The AI component considerably improves the system’s active surveillance capabilities, reaching a 94% fraud detection. Unlike traditional static verification systems, AIVS continuously examines academic data patterns to provide dynamic protection against evolving fraud techniques.

The low false positive rate demonstrates the system’s ability to discriminate between true academic improvement and anomalies, reducing the need for manual investigations. Verification time analysis adds to the practicality of AIVS. An 85% reduction in verification time compared to manual approaches shows that universities, companies, and credential verification services could profit greatly from implementing AIVS for fast, dependable validations. Immediate verification saves time, improves user experience, and decreases administrative workload.

Furthermore, the high satisfaction ratings among students 92%, instructors, and administrators indicate excellent user acceptance, which is frequently a significant obstacle to technological adoption in educational settings.

The benchmarking results presented in Table 3 indicate that although established systems like Turnitin, GPTZero, and Copyleaks exhibit robust baseline performance, the proposed model achieves superior accuracy and reduced false negative rates, which is crucial for identifying subtle anomalies in academic texts. Nonetheless, benchmarking reveals constraints: all systems, including ours, demonstrate significant false positive rates, which may disproportionately impact non-native English speakers. This underscores the necessity for human supervision and meticulous interpretation of automated outcomes in contexts of academic integrity.

The findings show that AIVS’s blockchain openness and AI-supported validation processes were well-received across all user segments. System resource consumption testing demonstrates that the AIVS architecture is extremely scalable. With CPU usage below 45% and minimal memory and blockchain storage overhead even under concurrent user loads, the system is ideal for deployment across medium to large academic institutions without requiring significant infrastructure investments.

Finally, the discussion of findings confirms that the AIVS framework achieves its intended goals by offering an efficient, scalable, and trustworthy system for protecting academic integrity. The findings suggest great potential for real-world application, particularly in situations where academic corruption risks are high and digital transformation of verification processes is urgently required. Our evaluation is slightly hindered by its dependence on a Proof-of-Authority (PoA) testnet. The Proof of Authority consensus mechanism provides accelerated confirmation times and reduced transaction costs; however, it is contingent upon an idealized structure with trustworthy validators, reduced decentralization, and a lower risk of attack in comparison to a public mainnet. The efficacy and reliability results from PoA may surpass those that can be achieved through practical decentralized implementation. In order to rectify this deficiency, additional testing is required under more difficult blockchain conditions with restricted resources. Table 4 shows blockchain storage and cost analysis, while Table 5 provides a summary of experimental evaluation results for the proposed AIVS framework.

Table 4

Table 4. Record size vs. blockchain overhead and gas cost.

Table 5

Table 5. Consolidated performance metrics of AIVS.

6.1 Discussion on AI detector bias

AI text detector bias hinders academic equity. These techniques commonly misclassify non-native English text as AI-generated due to simplified lexical and syntactic structures, according to a recent study. Liang et al. (2023) showed that widely used GPT detectors disproportionately identify non-native writing as machine-generated, but modest prompt-level alterations can pass. This shows unfair punishment and weakness. Commercial developers attempt to address these criticisms. Internal evaluations by Pratama (2025) showed no statistically significant bias against English language learners, with less than 1% false positives at specific thresholds. Subgroup discrepancies and generalizability remain issues after these results were not independently confirmed. Fiedler et al. (2025) noted that using such tools by default without openness around techniques and subgroup breakdowns risks compromising due process.

These dangers are heightened by independent investigations. The Markup reported that AI detectors disproportionately highlighted international students, accusing them of cheating based on single detection results. The NIU Center for Innovative Teaching and Learning (Weber et al., 2024) called detectors “an ethical minefield,” warning that even minor false-positive rates can disproportionately affect thousands of students in large classes.

Recent empirical research strengthens evidence. Hao and Angwin (2023) examined several detection algorithms and discovered accuracy and bias trade-offs: tighter thresholds improved average accuracy but worsened subgroup differences, especially among English language learners. In sophisticated academic settings, where style and aim matter. Oztas et al. (2024) found that human judgment is more reliable than machine detectors. These findings support criticism that detectors lack explainability, making their conclusions hard to dispute.

In addition, Turnitin (2023) found that prompt-level adversarial tactics can bypass detectors with modest text changes like synonym replacement or phrase rearrangement, reducing scale reliability. With tens of millions of flagged submissions (Vanderbilt University, 2023), even rare errors can erode student trust and academic integrity systems. The literature indicates that AI text identification should not be utilized in isolation for situations of academic dishonesty. Outputs should be seen as provisional indicators necessitating draft histories, oral defenses, or process-oriented evaluations. Academic equity necessitates transparent reporting of subgroup performance, human oversight, and regulatory protection. Future research will enhance validation by testing larger datasets and making direct comparisons with current verification platforms, including centralized university record systems and alternative blockchain-based credential models (e.g., Blockcerts, EduCTX). This will facilitate the quantification of AIVS’s performance enhancement in practical academic settings.

6.2 Discussion on scalability and cost implications

An essential consideration in assessing the practical viability of AIVS is its implementation cost and scalability on the Ethereum mainnet. The execution of smart contracts on Ethereum incurs transaction (gas) fees that fluctuate according to network congestion. Table 6 delineates the projected gas expenses for fundamental AIVS tasks, which include record insertion, verification, and audit logging across three scenarios: low, average, and high congestion. These figures are based on current Ethereum gas fee data and modified for standard contract complexity. The results demonstrate that whereas individual transactions are economical during low congestion, expenses escalate considerably under high usage conditions. In extensive deployments with thousands of records, the total cost may become excessive only on the Ethereum mainnet.

Table 6

Table 6. Estimated gas costs for AIVS smart contract functions on ethereum mainnet.

Layer-2 scaling methods, such as Optimistic Rollups like Optimism or Arbitrum, and sidechains such as Polygon, offer a viable answer to this issue. These systems aggregate several transactions off-chain and periodically submit summaries to Ethereum, decreasing per-transaction expenses by 50%–95%. Nevertheless, these methodologies have trade-offs, such as dependence on rollup operators, supplementary bridging mechanisms, and possible delays in transaction finality.

A hybrid architecture is thus advised: core credential commitments should be held on Ethereum for optimal security and auditability, while high-frequency verification and anomaly detection processes are to be transitioned to Layer-2 networks. This architecture harmonizes openness, security, and economic viability, guaranteeing that AIVS sustains its practicality at both national and international levels.

6.3 Validation using real institutional data and pilot deployment plan

The current assessment of AIVS utilized synthetic datasets to guarantee repeatability and mitigate risks; the subsequent phase of this research will entail controlled validation with authentic academic data in partnership with institutions in Kazakhstan, including the International Information Technology University (IITU), Satbayev University, and IT Astana University.

6.3.1 Data privacy and ethical compliance

Before executing real-world testing, the participating universities will provide anonymized and encrypted datasets comprising academic transcripts, course performance records, and thesis metadata. All data management will comply with the law of the Republic of Kazakhstan on personal data and its protection (2013) and any institutional privacy standards. Personally identifiable information (PII) will be removed or replaced with hashed identifiers, while access to original records will be restricted to authorized institutional personnel. Furthermore, blockchain transactions would contain only cryptographic hashes, excluding raw data, ensuring full compliance with data protection requirements akin to the EU GDPR. Ethical approval will be obtained from the Research Ethics Committee of each university before data exploitation.

6.3.2 Pilot deployment strategy

The pilot implementation will follow a three-phase roadmap:

⁃ Sandbox phase will be used, which consists of internal testing in which smart contracts, anomaly detection thresholds, and data flow integration with existing learning management systems are calibrated using anonymized academic records from a single department.

⁃ Controlled institutional pilot phase will be employed where a limited-scale deployment across multiple departments will be conducted that enables actual faculty submissions, student access, and employer verifications. This phase will gather feedback regarding the efficiency of the process, the veracity of the data, and the usability of the system.

⁃ Full-scale implementation will be executed, transitioning gradually to live institutional deployment, with interoperability testing undertaken among universities via platforms provided by the Ministry of Science and Higher Education.

6.3.3 Prospects for future integration

The proposed AIVS framework could be strategically integrated with the National digital diploma repository, which is administered by the Ministry of Education and Science of the Republic of Kazakhstan, following successful pilot validation. A unified, blockchain-based academic verification infrastructure would be established through this integration, which would facilitate secure, real-time credential authentication across universities, employers, professional licensing bodies, and accreditation authorities. This interoperability would expedite the recognition of academic qualifications on both a domestic and international scale, reduce the likelihood of document falsification, and eradicate redundant verification procedures. Additionally, the Ministry’s long-term vision for transparent, data-driven governance in higher education would be promoted by the national integration of AIVS, which is consistent with Kazakhstan’s Digital Transformation Strategy 2025 and the Anti-Corruption Program in Education. The system would establish a precedent for regional adoption across Central Asian educational systems, foster public trust, and reinforce institutional accountability by embedding immutable credential records on a distributed ledger.

Kazakhstan’s transition from a prototype to a nationally integrated infrastructure would provide compelling evidence of AIVS’s real-world viability, scalability, and socio-educational impact, which would mark a significant milestone in the country’s efforts to combat corruption and promote integrity within academic ecosystems.

7 Conclusion and future works

In this section, we summarize the major findings of the study and outline potential future directions to enhance the proposed academic integrity verification system further.

7.1 Conclusion

The academic integrity verification system is a cutting-edge framework that uses blockchain technology and artificial intelligence to prevent academic achievements from being modified or fraudulently earned. This study conceptualized and built the system. We have demonstrated, through its development and experimental evaluation, that the AIVS effectively addresses this challenge by mitigating the vulnerabilities inherent in traditional academic record management. This is accomplished by combining the immutability of the blockchain with the proactive anomaly-detecting capabilities of AI. All our experiments have demonstrated that the underlying hypothesis is valid. AIVS achieved zero occurrences of tampering, a 94% fraud detection rate, an overall model accuracy of 95%, and an 85% reduction in the time necessary to validate academic records when compared to conventional approaches. Students, instructors, and administrators all praised the system, showing that it was well-accepted and useful. The results show that combining blockchain technology and artificial intelligence can dramatically improve the safety, transparency, and efficacy of academic verification processes.

The AIVS framework establishes a new standard for higher education governance by protecting academic integrity, lowering administrative constraints, and promoting the reliable and timely verification of credentials. The AIVS framework has the potential to improve education in Kazakhstan while also promoting a global push to create more accessible, safe, and dependable academic environments. This is due to the introduction of a new criterion for academic honesty. The proposed AIVS framework demonstrates significant viability and encouraging outcomes at the prototype stage. However, its nationwide incorporation into the National digital diploma repository is a long-term objective. Future implementation will necessitate substantial coordination with the Ministry of Education and Science, thorough pilot assessments, and policy alignment with Kazakhstan’s digital transformation strategy 2025. The results of this investigation thus signify a proof-of-concept phase for the creation of a scalable, nationally integrated academic verification platform.

7.2 Future work

Future studies will focus on adding universities. A standardized decentralized infrastructure simplifies university academic verification. AIVS can improve national digital identification systems by verifying users and enhancing authentication. To promote national and global decentralization, transparency, and auditability, the AIVS platform will be tested on the Ethereum mainnet or Hyperledger Fabric. Continuous learning improves AI modules. With this, anomaly detection systems might learn new fraud patterns without operator training. Plans include student and employer mobile apps. These apps will make academic record verification easier for students on the go and globalize the system. More stakeholder groups will undergo rigorous usability testing to improve the system interface. AIVS will remain user-friendly in academic and professional environments. Last, energy efficiency in blockchain operations will be a focus due to energy usage. To lessen the environmental impact of large-scale installations, we will investigate PoA or Layer-2 scaling. These paths can make the AIVS a strong, versatile, and globally relevant academic integrity verification system.

Data availability statement

The data analyzed in this study is subject to the following licenses/restrictions: We are still working to generate more results so we will make available publicly later. Requests to access these datasets should be directed to Will be available on IT Astana.

Author contributions

AR: Methodology, Validation, Conceptualization, Software, Writing – review and editing, Visualization, Writing – original draft, Formal Analysis, Data curation. SA: Funding acquisition, Writing – review and editing, Methodology, Conceptualization, Writing – original draft, Resources. GM: Software, Investigation, Writing – review and editing, Data curation, Validation. OU: Writing – review and editing, Resources, Validation, Funding acquisition, Software. MF: Writing – review and editing, Writing – original draft, Formal Analysis, Validation. YK: Resources, Software, Funding acquisition, Writing – review and editing, Data curation. SK: Investigation, Writing – review and editing, Validation, Data curation, Formal Analysis. KA: Writing – review and editing, Validation, Investigation, Software, Data curation, Visualization.

Funding

The authors declare that financial support was received for the research and/or publication of this article. This research was conducted under project BR24993014 on developing an intelligent anti-corruption system for validating educational achievements and official documents in Kazakhstani universities, implemented by the Institute of Information and Computational Technologies.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The authors declare that no Generative AI was used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbloc.2025.1683522/full#supplementary-material

References

Alsulami, B. M. (2024). Blockchain integrated data processing model for enabling security in education 4.0. Fusion Pract. and Appl. 14 (1), 190–220. doi:10.54216/fpa.140116

CrossRef Full Text | Google Scholar

Anwar, F., Khan, B.U. I., Kiah, M. L. B. M., Abdullah, N. A., and Wen Goh, K. (2022). A comprehensive insight into blockchain technology: past development, present impact and future considerations. Int. J. Adv. Comput. Sci. Appl. 13 (11). doi:10.14569/ijacsa.2022.01311101

CrossRef Full Text | Google Scholar

Any, B., Ramadhan, T., Nabila, E. A., and Efa Ayu Nabila, (2024). Decentralized academic platforms: the future of education in the age of blockchain. Blockchain Front. Technol. 3 (2), 112–124. doi:10.34306/bfront.v3i2.467

CrossRef Full Text | Google Scholar

Balobaid, A. S., Yasamin, H. A., Ali, H. F., Jamal, N. H., et al. (2023). Modeling of blockchain with encryption based secure education record management system. Egypt. Inform. J24 (4): doi:10.0411.10.1016/j.eij.2023.100411

CrossRef Full Text | Google Scholar

Boulet, P., de Coëtlogon, P., Thébault, É., van de Velde, P., and Murie, C. (2025). Digital transformation of certificates issued by universities for European competitiveness. Université de Lille, Lille, France. PhD Dissertion.

Google Scholar

Chaudhari, S., and Shirole, M. (2025). Blockchain-driven academic learning record management in higher education: a comprehensive review of methodologies, applications, benefits, and challenges. SN Comput. Sci. 6 (5), Article 427. doi:10.1007/s42979-025-03952-z

CrossRef Full Text | Google Scholar

Chen, Z., Jiang, H., You, J., Wang, X., and Sun, P. Z. H. (2024). RFID lightweight authentication mechanism for smart factories based on blockchain. IEEE J. Radio Freq. Identif. 8, 19–30. doi:10.1109/jrfid.2024.3356194

CrossRef Full Text | Google Scholar

Chinnasamy, P., Subashini, B., Ayyasamy, A. R. K., Kiran, A., Pandey, D. B. K., Pandey, D., et al. (2025). Blockchain based electronic educational document management with role-based access control using machine learning model. Sci. Rep. 15, 18828. doi:10.1038/s41598-025-99683-5

PubMed Abstract | CrossRef Full Text | Google Scholar

Davletbayeva, Z., Bekish, U. A., Zagrebin, A. V., and Raisova, L. F.(2025).“Enhancing the role of public councils as an anti-corruption mechanism: the case of Kazakhstan.”9. doi:10.51889/2959-6270.2025.89.1.009

CrossRef Full Text | Google Scholar

Eaton, S. E., and Carmichael, J. J. (2023). Fake degrees and credential fraud, contract cheating, and paper mills: overview and historical perspectives. Fake Degrees Fraudulent Credentials High. Educ., 1–22. doi:10.1007/978-3-031-21796-8_1

CrossRef Full Text | Google Scholar

Fatima, B., and Senthilkumar, P. (2025). Leveraging zero-knowledge proofs for privacy-preserving blockchain transactions. ICICC 2024. doi:10.2139/ssrn.5239705

CrossRef Full Text | Google Scholar

Fiedler, A., Jensen, L. R., and Kovač, M. (2025). Do humans identify AI-Generated text better than machines?. Comput. and Educ. X Real.

Google Scholar

Gräther, W., Kolvenbach, S., Ruland, R., Schütte, J., Torres, C., and Wendland, F. (2018). “Blockchain for education: lifelong learning passport,” in Proceedings of 1st ERCIM blockchain workshop 2018 (European Society for Socially Embedded Technologies EUSSET). Amsterdam, Netherlands

Google Scholar

Gupta, N., Jain, R., and Chauhan, P. S. (2024). “Reforming the education system with blockchain technology,”Int. Conf. Integr. Circuits, Commun. Comput. Syst. (ICIC3S), 1. 1–5. doi:10.1109/icic3s61846.2024.10603321

CrossRef Full Text | Google Scholar

Hao, K., and Angwin, J. (2023). AI detection tools falsely accuse international students of cheating. Markup.

Google Scholar

Ishkov, A., and Krupnov, Y. (2024). “RETRACTED: application of blockchain technologies for creation of innovative ecosystem in higher education,”E3S Web Conf., 549. 08031. doi:10.1051/e3sconf/202454908031

CrossRef Full Text | Google Scholar

Kamarudin, N. H., Suhaimi, N. H. S., Rashid, F. A. N., Khalid, M. N. A., and Ali, F. M. (2024). Exploring authentication paradigms in the internet of things: a comprehensive scoping review. Symmetry EDP Sciences: Les Ulis, France16 (2), 171. doi:10.3390/sym16020171

CrossRef Full Text | Google Scholar

Kerimkulova, S., Goodman, B., and Aitzhanova, K. (2023). “EMI in higher education in Kazakhstan: policy supports and challenges,” in Policies, politics, and ideologies of english-medium instruction in Asian universities (Routledge Abingdon (Oxfordshire), United Kingdom), 15–32.

CrossRef Full Text | Google Scholar

Khan, A., Ali Laghari, A., Ahmed Shaikh, A., Bourouis, S., Mamlouk, A. M., and Alshazly, H. (2021). Educational blockchain: a secure degree attestation and verification traceability architecture for higher education commission. Appl. Sci. 11 (22), 10917. doi:10.3390/app112210917

CrossRef Full Text | Google Scholar

Kustandi, C., Bakhar, M., Pangestu Hadinigrum, L., Oktarina, N., and Nugraha, M. A. P. (2024). Utilizing blockchain technology for the advancement of education. AIP Conf. Proc., AIP Publishing (or American Institute of Physics via AIP Publishing) Melville, NY, US 3098 1 020012. doi:10.1063/5.0233074

CrossRef Full Text | Google Scholar

Lavin, R., Liu, X., Mohanty, H., Norman, L., Zaarour, G., and Krishnamachari, B. (2024). A survey on the applications of zero-knowledge proofs. AIP Publishing (or American Institute of Physics via AIP Publishing). Melville, NY, US arXiv preprint.

Google Scholar

Li, Z. Z., Liu Joseph, K., Yu, J., and Gasevic, D. (2022). “Blockchain-based solutions for education credentialing system: Comparison and implications for future development,” in 2022 IEEE international conference on blockchain (blockchain) (IEEE), 79–86.

Google Scholar

Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., and Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns 4 (7), Article 100779. doi:10.1016/j.patter.2023.100779

PubMed Abstract | CrossRef Full Text | Google Scholar

Lu, M., Han, Z., Rao, S.X., Zhang, Z., Yang, Z., Shan, Y., et al. (2022). “Bright-graph neural networks in real-time fraud detection,” in Proceedings of the 31st ACM international conference on information and knowledge management, 3342–3351.

Google Scholar

Nadeem, N., Hayat, M. F., Qureshi, M. A., Majid, M., Nadeem, M., and Janjua, J. (2023). Hybrid blockchain-based academic credential verification system (B-ACVS). Multimedia Tools Appl. 82 (28), 43991–44019. doi:10.1007/s11042-023-14944-7

CrossRef Full Text | Google Scholar

Oztas, B., Cetinkaya, D., and Adedoyin, F. (2024). AI detectors: an ethical minefield. NIU Center for Innovative Teaching and Learning. Institutional Report.

Google Scholar

Panzabekova, A., Fazylzhan, D., and Imangali, Z. (2024). Analyzing the relationship between corruption and socio-economic development in Kazakhstan’s regions. R-Economy NIU Center for Innovative Teaching & Learning,DeKalb, Illinois, USA 10 (4), 455–474. doi:10.15826/recon.2024.10.4.028

CrossRef Full Text | Google Scholar

Pratama, A. R. (2025). The accuracy-bias trade-offs in AI text detection tools and their impact on fairness in scholarly publication. PeerJ Comput. Sci. 11, e2953. doi:10.7717/peerj-cs.2953

PubMed Abstract | CrossRef Full Text | Google Scholar

Ramasamy, L. K., and Khan, F. (2024). “Secure and transparent educational data record-keeping with blockchain,” in Blockchain for global education (Cham: Springer Nature Switzerland), 147–164.

CrossRef Full Text | Google Scholar

Razaque, A., Du, K., Zhu, X., Wanyue, L., Hani, Q. B., and Khan, M. J. (2018). “Survey: wildlife trade and related criminal activities over the internet,” in 2018 IEEE Long Island systems, applications and technology conference (LISAT) (IEEE), 1–6.

Google Scholar

Razaque, A., Frej, M. B. H., Bektemyssova, G., Amsaad, F., Almiani, M., Aziz, A., et al. (2022). Credit card-not-present fraud detection and prevention using big data analytics algorithms. Appl. Sci. 13 (1), 57. doi:10.3390/app13010057

CrossRef Full Text | Google Scholar

Rostamkolaei Motlagh, S. M., Pahl, C., Barzegar, H. R., and El Ioini, N. (2025). A technology review of zero knowledge proof techniques. IoTBDS 2025, 245–253. doi:10.5220/0013269600003944

CrossRef Full Text | Google Scholar

Rustemi, A., Dalipi, F., Atanasovski, V., and Risteski, A. (2023). A systematic literature review on blockchain-based systems for academic certificate verification. IEEE Access 11, 64679–64696. doi:10.1109/access.2023.3289598

CrossRef Full Text | Google Scholar

Sharwani, A. E., and Melo, R. (2024). Blockchain-enabled academic credential verification in the USA. Adhyayan A J. Manag. Sci. 14 (02), 31–41. doi:10.21567/adhyayan.v14i2.07

CrossRef Full Text | Google Scholar

Tong, F., Chen, X., Wang, K., and Zhang, Y. (2022). CCAP: a complete cross-domain authentication based on blockchain for internet of things. IEEE Trans. Inf. Forensics Secur. 17, 3789–3800. doi:10.1109/tifs.2022.3214733

CrossRef Full Text | Google Scholar

Turkanović, M., Hölbl, M., Košič, K., Heričko, M., and Kamišalić, A. (2018). EduCTX: a blockchain-based higher education credit platform. IEEE access 6, 5112–5127. doi:10.1109/access.2018.2789929

CrossRef Full Text | Google Scholar

Turnitin (2023). New research: turnitin’s AI detector shows no statistically significant bias against ELLs. Turnitin, LLC Oakland, California, USA.

Google Scholar

Vanderbilt University (2023). Guidance on AI detection and why we’re disabling turnitin’s AI detector. Vanderbilt University, Center for Teaching Nashville, Tennessee, USA.

Google Scholar

Vevera, A. V., and Botezatu, U. E. (2024). “Exploring the impact of blockchain in educational settings: a comprehensive analysis,” in INTED2024 proceedings (IATED Academy Valencia, Spain), 5403–5408.

CrossRef Full Text | Google Scholar

Weber, S., Müller, T., and Scholz, H. (2024). Adversarial strategies for bypassing AI text detectors in academic writing. J. Educ. Technol. Res.

Google Scholar

Widayanti, R., Purnama Harahap, E., Lutfiani, N., Putri Oganda, F., and Manik, I. S. P. (2021). The impact of blockchain technology in higher education quality improvement. J. Ilm. Tek. Elektro Komput. Dan. Inf. 7, 207–216. doi:10.26555/jiteki.v7i2.20677

CrossRef Full Text | Google Scholar

Xu, Y. (2024). Development of blockchain-based academic credential verification system. Open Access Libr. J. 11 (9), 1–19. doi:10.4236/oalib.1112130

CrossRef Full Text | Google Scholar