Self-regulated learning is an empirically derived concept that aids in managing and controlling the cognitive, affective, behavioural, and emotional state of the learners. The three cyclic phases of SRL (Zimmerman, 2002), involving the forethought phase for goal orientation, the self-monitoring phase for monitoring one’s performance (Osakwe et al., 2024; Hashem, 2021) and the final self-evaluation phase for self-reflection (Alqahtani, 2024), make it an iterative and dynamic learning process that keeps improving with time. However, incorporating SRL within a flipped classroom requires equal distribution of responsibility among instructors and learners.

Learners must engage independently (Goel, 2024) with the learning content, plan their schedules effectively, and be metacognitively aware in the pre-class phase (Ouyang et al., 2023). Proper strategic planning, problem-solving skills, and collaboration with peers are the prerequisites for the in-class phase to be beneficial. Further, the post-class hours focus on self-reflection (Alqahtani, 2024), which helps learners to reinforce their learning patterns (Panadero, 2017; Gligorea et al., 2023) and revise their learning pathways. Most experimental research supports the effectiveness of SRL in achieving learning outcomes and in supporting more persistent performance in blended learning environments (Elbourhamy et al., 2023). Evidence shows a deeper understanding of the concepts, stronger metacognitive abilities (Samadi et al., 2024), and high retention capabilities (Dignath et al., 2023; Wang et al., 2021) among learners who were successful enough to implement the cyclic phases in their learning process. Nevertheless, the lack of self-regulation skills in learners, lack of motivational factors, and absence of self-autonomy reported in a few studies demonstrate that the successful implementation of a flipped classroom is not only determined by the instructional design strategies but also by learners’ self-determination and regular monitoring of the learning process (Cheng et al., 2024).

Therefore, the SRL integration into a flipped classroom setup remains operational only through the incorporation of activities, tools, and videos without much theoretical grounding, as evident in previous works. Furthermore, despite the significance in enhancing learner’s interest, engagement, and resilience, the role of emotional and social dimensions in an SRL framework remains unexplored. This barrier creates future opportunities to explore and integrate the social and emotional aspects of learning into an SRL-driven flipped classroom.

In a blended or hybrid learning environment, it is always crucial to understand learners’ emotions, motivation and engagement levels, metacognitive abilities, and attitudes towards the learning process (Zheng et al., 2023). Moreover, learning in a digital environment does not constitute only the cognitive behaviour of the learner but also the social and emotional wellbeing (Su and Fung, 2024) of the learners. To address this conceptual gap in the prior works, this study employs the social–emotional intelligence (SEI), signifying both social intelligence (SI) and emotional intelligence (EI) as an integral construct of the SRL-oriented flipped classroom methodology. Interpersonal competencies like effective communication, resolving conflicts of interest, collaboration, and empathy (Akti Aslan, 2022) constitute social intelligence. In contrast, intrapersonal skills such as emotional awareness, resilience, stress management, and emotional regulation (Patil et al., 2023) fall under the category of emotional intelligence. Apart from these competencies, common emotions such as happiness, enthusiasm, enjoyment, anxiety, depression, and frustration play a significant role in shaping learners’ mental abilities (Du et al., 2023).

Furthermore, emotional stability and resilience (Goleman, 1998; Parker et al., 2004) are fostered through the integration of social–emotional intelligence (SEI) with self-regulated learning (SRL), which are essential in self-directed learning contexts (Herut et al., 2024). To enable learners to manage stress and anxiety, remain consistent during challenging tasks, and sustain self-motivation during the learning process, self-awareness of emotions and effective regulation become essential. Similarly, facilitation of social intelligence via collaborative and group tasks, peer interactions, and sharing productive feedback is the key to the in-class phase of the flipped classroom environment. Recent studies indicate that learners with higher emotional intelligence significantly show improved performance with better self-regulatory behaviour and perseverance (Silva et al., 2023).

Nevertheless, another significant aspect in proximity to SEI is the engagement metric, which is contemplated in this study as a multifaceted construct comprising cognitive, emotional, and behavioural components. Cognitive engagement denoted the learning strategies adopted by learners, emotional engagement recorded common feelings and affective state of learners, and behavioural engagement reflected the participation levels, interactions, and task completion rates (Fredricks et al., 2019; Bond et al., 2020). In this context, terms such as ‘student engagement’ and ‘learner engagement’ are used across formal, self-directed, and self-regulated learning contexts (Henrie et al., 2023). Following the engagement metric, ‘affect’ is another such construct representing the momentary emotions and feelings experienced by the learners during the learning process. Whereas affective analytics refers to the systematic analysis of affective states, functioning as a supportive measure to observe, evaluate and interpret the emotional intelligence of the learners (D’Mello and Graesser, 2021; Joksimović et al., 2023). Despite the growing significance of affective analytics in digital learning, only a few studies on the flipped classroom methodology have attempted to incorporate affective data with the SRL phases, thus leaving an empirical gap and opportunities for future researchers to address this challenge (Khalil and Ebner, 2023).

Digital learning platforms have increasingly enabled self-regulated learning by externalising learners’ cognitive, behavioural, and emotional processes, while technology can function as a metacognitive scaffold, providing learners with timely feedback on goal progress, engagement patterns, emotional states, and learning outcomes (Heikkinen et al., 2023; Silva et al., 2023). Such productive feedback helps transform passive learning processes into insightful actions to support the monitoring and reflection phases of SRL. Moreover, in a flipped classroom context, educational technologies facilitate SRL in multiple ways. While pre-class preparation is achieved via incorporating the usage of Learning management systems and collaboration platforms, support planning and resource management, digital tools enable peer interaction, formative assessment, and immediate feedback during in-class sessions, reinforcing self-monitoring and strategy adjustment (Ouyang et al., 2023; Osakwe et al., 2024). Further, the post-class reflective journals and portfolio systems promote self-evaluation and longitudinal reflection, helping learners track their progress over time (Cheng et al., 2025).

Research in learning analytics demonstrates that data-informed feedback and coaching can enhance learners’ self-awareness, strategic regulation, and academic persistence (Heikkinen et al., 2023; Khalil et al., 2024). Nevertheless, the current literature predominantly addresses technological support for SRL in a disjointed fashion. Numerous studies focus on cognitive analytics or performance prediction, while others examine engagement or emotional variables independently. There remains limited empirical work that integrates SRL processes, social–emotional intelligence, engagement analytics, and machine-learning-driven (Cardona et al., 2023) modelling within authentic flipped classroom implementations, particularly in engineering and computer science education (Wong et al., 2025). Therefore, there is a distinct need to develop a holistic framework that integrates pedagogical strategies, SRL theory, affective and engagement analytics, and advanced data-driven methods to comprehend how technology can support the complete SRL cycle, including its emotional and social dimensions. Addressing this gap creates opportunities for a more thorough investigation of the managerial skills of students to manage their learning in a technologically advanced flipped context on a cognitive, social, and emotional level. Educational platforms like Microsoft Teams and journaling applications have been implemented to enhance learner participation and engagement, along with positive feedback (Li and Wang, 2023). The essential aspects of self-regulated learning (van Alten et al., 2020), namely, team collaboration, workspace sharing, and self-monitoring of progress, are facilitated through these platforms. Nonetheless, there is an insufficient body of research connecting self-regulated learning techniques to social and emotional intelligence within a digital environment, hence affecting academic progression.

The current literature highlights specific inadequacies listing the first one as a deficiency of empirical research about the integration of self-regulated learning (SRL) and social–emotional intelligence (SEI) in flipped classrooms that employ digital resources, secondly the insufficient use of affective and engagement analytics in assessing the efficacy of self-regulated learning (SRL) (Psathas et al., 2023), and followed by insufficient application of machine learning for predicting student engagement (Arizmendi et al., 2023; Azizah et al., 2024) and academic success in hybrid learning contexts. The subsequent research questions (RQs) are constructed to address the highlighted deficiencies in the study.

This study addresses these shortcomings by aligning pedagogical methods with technological advancements, fostering psychological resilience, and equipping individuals to address contemporary practical issues.

This study employs a mixed-methods quasi-experimental approach to evaluate the outcomes of a flipped SRL model on learners’ performance and engagement levels, as well as their social and emotional intelligence. The proposed model is primarily theory-driven by Zimmerman’s cyclic SRL model, complemented with social and emotional dimensions of learning.

Zimmerman visualised SRL as a 3-phase cyclic process comprising forethought, self-monitoring, and self-reflection, occurring iteratively. This iterative cycle closely aligns with the flipped classroom methodology, which spreads the learning process across 3 different phases of pre-class, in-class, and post-class sessions. Therefore, the proposed framework depicted in Figure 1 implements the three SRL phases incorporated within the flipped classroom environment as follows.

The proposed framework integrates social and emotional intelligence as a moderator, along with the cognitive and behavioural dimensions. Therefore, this flipped SRL model combines the theoretical grounding of SRL, the structure of the flipped classroom and technology-enabled learning analytics as illustrated in Figure 1. The success of this integrated model is supported by prior evidence demonstrating that learners with better emotional awareness, management, and regulation tend to have higher engagement levels and sustained academic performance.

A total of 96 undergraduate students enrolled in a third-year Database Management Systems (DBMS) course in a Computer Science and Engineering programme participated in this study. Of the enrolled cohort, complete datasets were available for all 96 participants across academic performance measures, Continuous Assessment Test 1 (CAT 1), Continuous Assessment Test 2 (CAT 2) and Final Assessment Test (FAT), self-regulated learning artefacts (journals and portfolios), engagement logs, and social–emotional intelligence surveys. No participants were excluded from the analysis, as all students met the inclusion criteria of course enrolment and completion of required learning activities. The sample consisted of students within a relatively homogeneous academic context in terms of programme level and disciplinary background. Participation was incorporated within the course as part of regular instructional activities; however, the use of data for research purposes followed institutional ethical guidelines. Descriptive statistics for key variables, including academic performance, SRL indicators, engagement metrics, and social–emotional intelligence measures, are reported in the Results section to provide transparency and enable assessment of potential bias and variability across the sample.

The study was conducted for third-year undergraduate students enrolled in the Database Management System course in the Computer Science and Engineering stream. The course followed a flipped classroom structure implemented with 3 SRL phases as given in Figure 2 and spanned across one semester.

The course began with an introduction to the SRL techniques, their principles, and their relevance to the course and the flipped learning methodology. Pre-class sessions included micro-learning content, relevant flipped lecture videos, digital resources, and formative assessments such as quizzes, which were shared on Microsoft Teams and other associated tools like Clockify and Evernote to encourage goal setting, planning, and the stimulation of prior knowledge. In-class sessions included group learning activities and discussions that focused on solving problems and getting immediate feedback from the instructors. These activities helped learners in self-monitoring their learning process. Digital journaling and e-portfolio development were involved as reflective practices (Robbins et al., 2020) in the post-class sessions, enabling learners to evaluate their emotional wellbeing and progress over time.

Numerous digital tools were incorporated in the course to facilitate the SRL model by mapping these tools explicitly with each phase of SRL. The central platform used to deliver content, conduct lectures, and interact with peers and facilitators was Microsoft Teams. To activate prior knowledge and engagement during the in-class activities, tools like Quizizz and Mentimeter were employed for formative assessments. Clockify was used to aid the learners in planning their tasks and maintaining time and schedules during rush hours, and to monitor the duration and time spent on each task. Additionally, the emotional states and feelings of the learners were recorded using the Reflect app during the semester. Furthermore, digital journaling was implemented using the Evernote application to record learner experiences, emotions, achievements and strategic plans, and E-Portfolio creation fostered self-evaluation and representation of the overall achievements and artefacts of learning, facilitating long-term reflection.

This subsequent subsection outlines the measurement instruments and their functions within the comprehensive proposed framework. The instruments were selected and modified to operationalise the study’s key constructs in the flipped classroom setting, aligning with the established theoretical frameworks. The multidimensional metrics of SRL were engineered to obtain both behavioural and perceptual dimensions of learning. Additionally, the reliability, construct validity, and the correspondence between theoretical definitions and empirical evidence are also emphasised in this sub-section, ensuring conceptual clarity and analytical rigour.

Internal consistency reliability (Cronbach’s α) was computed for each domain to evaluate internal coherence among items. Consistent with psychometric standards, reliability indices do not constitute evidence of construct validity; therefore, the adapted SEI instrument should be interpreted as a theoretically grounded measure rather than a fully validated scale. The complete item set and domain structure are provided in Appendix C for transparency.

Table 2 displays the descriptive statistics for the principal variables examined in this study, incorporating assessment scores, reflective learning indicators, engagement metrics, and social–emotional intelligence (SEI) scores for the 96 samples considered. The results from the baseline (CAT 1), mid-semester (CAT 2), and final assessment (FAT) exhibit moderate variability among students, with FAT reflecting a higher mean than both CAT 1 and CAT 2, indicating an enhancement in academic performance over the semester. Journaling and portfolio activity scores demonstrated significant variability, indicating disparities in students’ involvement with reflective learning tasks and course-based assignments. SEI scores exhibit less variability, signifying a more uniform distribution of social–emotional abilities among the cohort. Indicators related to engagement further corroborate these patterns. Additionally, inconsistencies in completing the given learning tasks within the course were indicated through moderate variability in completion rates, whereas feedback scores showed lesser variance, signifying a consistent quality of feedback and student engagement throughout the group.

The supplementary data facilitating the interpretation of the descriptive statistics are provided in Appendix D, including supportive engagement analytics and platform interaction visualisations obtained through Microsoft Teams and Power BI applications. The revised SEI questionnaire domains and their corresponding statistical item mappings are comprehensively detailed in Appendix E, while the feedback of students after the semester has concluded is presented in Appendix F.

Regression and mediation analyses were performed to examine associative and explanatory relationships among self-regulated learning, learner engagement, social–emotional intelligence, and students’ academic performance across different exams (CAT 1, CAT 2, and FAT). These analyses do not claim causal effect experimentally, but rather test theoretically informed pathways and directional assumptions.

A simple linear regression model revealed that the SRL Score was a strong predictor of learner engagement (β = 1.77, p < 0.001), accounting for 40.4% of the variance in engagement (R² = 0.404). The second regression model indicated that engagement strongly predicted FAT scores (β = 3.73e-07, p < 0.001, R² = 0.288), with a modest effect, suggesting that engagement is a contributing factor rather than the exclusive driver of academic achievement. Another regression model confirmed a significant direct effect (β = 1.359e-07, p < 0.001, R² = 0.345) of SRL on academic performance, indicating that SRL influences FAT both directly and indirectly via engagement. Consequently, the current analyses demonstrate the statistical relationships via associations, temporal order between SRL, CAT 2, and FAT results, and nonspuriousness as far as previous theoretical and empirical work supports it. Given the sample size (N = 96), effect estimates should be interpreted cautiously, particularly for mediation pathways. These restrictions justify the exploration of the mediation pathways as theoretically viable explanatory mechanisms rather than conclusive causal claims.

Additionally, the illustration in Figure 3, depicting a scatter plot with a regression line, represents the association between reflective SRL activities, like journaling and portfolio management, with FAT scores. A clear positive trend is identified, indicating that students with higher self-regulated learning (SRL) engagement patterns tend to produce superior performance in a flipped classroom environment. Nevertheless, despite lower self-regulated learning skills, a few students excelled, highlighting the influence of external factors. Overall, the Variability observed in the scatter plot reflects heterogeneity in SRL application across learners.

The extent to which machine learning models can predict academic performance outcomes (Hussain and Khan, 2023) and classify learners based on their engagement levels, using derived SRL-metrics, social–emotional factors, and engagement indicators, is addressed by RQ2.

Regression-based ensemble models, namely, Random Forest (RF) (Badal and Sungkur, 2023), Gradient Boosting Machine (GBM), and XGBoost, were evaluated (Alhazmi and Sheneamer, 2023) using Mean Squared Error (MSE), R², Adjusted R², and Mean Absolute Percentage Error (MAPE) metrics. Among these models, XGBoost outperformed the other models with the lowest mean squared error (12.82) and the highest R² value (0.83) as given in Table 3, closely followed by Gradient Boosting with an R² of 0.82, while Random Forest had the lowest predictive accuracy (R² = 0.74). A Five-fold cross-validation was applied to mitigate overfitting risk, and the feature importance analysis identified SRL composite score and portfolio-related indicators as the most influential predictors. Given the modest sample size (N = 96), predictive performance should be interpreted as exploratory rather than generalisable. However, model complexity was balanced against sample size to reduce variance inflation.

Students were classified into performance-based groups based on academic achievement in FAT using Support Vector Machine (SVM) and Decision Tree (DT) models (Arashpour et al., 2023). The SVM model outperformed the DT model in key performance metrics across accuracy (75% vs. 55%), precision, recall, F1-score, and MCC, as given in Table 4, indicating greater stability and discriminative capacity.

These findings suggest that the SRL-related features and the engagement metrics can be leveraged to model and forecast learner’s performance outcomes with reasonable accuracy, hence validating the effectiveness of learning analytics in an SRL-integrated flipped classroom environment.

RQ3 investigated the associations and relationships among Social–Emotional Intelligence, Emotional States, and SRL outcomes using the Reflect app and SEI questionnaires to explore students’ emotional conditions and interactions. The analysis of emotional clusters revealed two dominant categories: Positive Emotion Cluster, characterised by calmness, focus and attentiveness, and the Negative Emotion Cluster associated with stress, anxiety, and loneliness. Designated as Whirlwind Emotions. About 36% of students belonged to the positive emotions cluster named Lake emotions, while merely 4% of the student population was categorised within the negative emotions cluster termed as Whirlwind emotions (Refer to Appendix D).

The majority group exhibited consistent engagement with higher participation levels in the self-regulated learning (SRL) activities. Additionally, the correlation analysis indicated moderate positive relationships among empathy, emotional regulation, adaptability, and academic performance (r = 0.64, p < 0.01), indicating that SEI coexist with sustained engagement and self-regulation in flipped classroom contexts. The results validate the third inquiry, highlighting the intricate interplay between emotional states, engagement, and performance outcomes in the context of SRL and higher education. Furthermore, the correlation heatmap presented in Figure 4 examines relationships among SRL indicators, SEI dimensions, engagement metrics, and performance outcomes. “Portfolio Performance Impact” and “Self-Regulation Effectiveness” showed strong positive associations (r = 0.94), in addition to “Cognitive Engagement” and “Self-Regulation Effectiveness” (r = 0.83).

Moreover, positive connections were identified between “CAT 2” scores with “Portfolio” management and “Self-Regulation Effectiveness,” and students with better initial scores showed lesser improvement over time, indicating negative associations between “Performance Growth Rate” and initial assessment scores. Thus, the heatmap illustrates significant interrelationships among self-regulation, cognitive engagement, and academic outcomes within the flipped SRL environment. Considering the sample size (N = 96), the correlation and clustering results should be interpreted cautiously. While the analyses reveal meaningful patterns linking social–emotional intelligence, engagement, and SRL outcomes, cluster stability and correlation estimates may vary in larger datasets. Therefore, these findings should be regarded as exploratory indicators requiring validation in broader samples.

Further, K-Means clustering was applied to group students as given in Figure 5 based on SRL scores and participation levels, identifying three engagement profiles. The Elbow method was applied to choose the number of clusters, and further, the three-cluster solution yielded a Silhouette Score of 0.42, indicating moderate cluster cohesion and acceptable separation. The clusters formed demonstrated that the high-engagement group showed consistently stronger SRL indicators, whereas the low-engagement group showed limited participation. Although cluster boundaries are not sharply defined, reflecting the continuous nature of learner engagement, the solution captures meaningful variability in SRL-related participation patterns.

The sample size (N = 96) provides adequate power for moderate regression effects but limits the detection of small effects. Mediation pathways mentioned in RQ1 should be interpreted cautiously due to sample size constraints. In RQ2, despite the cross-validation of the machine learning models, predictive stability may vary in larger populations. Furthermore, the clustering solutions given in RQ3 should be considered exploratory due to potential sensitivity to sample size.

The findings of this study align with and extend prior research highlighting the relationship between self-regulated learning (SRL), learner engagement, and academic performance in flipped and technology-enhanced learning environments. Consistent with earlier studies, learners who demonstrated stronger planning, monitoring, and reflection behaviours also exhibited higher engagement levels and improved academic outcomes (Dignath et al., 2023; Samadi et al., 2024). These patterns support the theoretical perspective that SRL operates as a cyclical process in which goal setting, strategic action, and reflection collectively shape learning outcomes (Zimmerman, 2013; Panadero, 2017).

The observed association between engagement indicators and academic performance further corresponds with engagement frameworks that conceptualise cognitive, behavioural, and emotional participation as key drivers of learning persistence and achievement (Fredricks et al., 2019; Henrie et al., 2023). In flipped classroom environments specifically, previous work has shown that structured pre-class preparation and in-class collaboration promoted deeper engagement and active knowledge construction (Sun et al., 2023; Samadi et al., 2024). The present findings therefore reinforce the interpretation that engagement functions as an important mechanism through which SRL-related behaviours are reflected in observable academic outcomes.

In addition, the association between social–emotional intelligence (SEI) indicators and reflective learning behaviours highlights the importance of emotional regulation and interpersonal awareness within self-regulated learning contexts. Prior research shows that learners with stronger emotional awareness and stress-management skills demonstrate greater persistence and adaptability while facing academic challenges (Silva et al., 2023; Su and Fung, 2024). Similarly, emotional intelligence has been correlated with enhanced collaborative learning and sustained motivation in technology-supported environments (Herut et al., 2024). The current findings, are a valuable addition to the expanding body of literature suggesting that SRL processes are not exclusively cognitive but are also influenced by engagement and affective dimensions of learning.

Furthermore, the application of predictive modelling and learning analytics in this study illustrated how behavioural indicators derived from digital learning environments provided additional insights into SRL processes. Present research in learning analytics emphasises that data from digital platforms reveal patterns of engagement, persistence, and strategic learning behaviours that are not easily observable through conventional assessment methods (Heikkinen et al., 2023; Khalil et al., 2024). The predictive models explored in this study should therefore be interpreted as exploratory analytical tools that help identify potential relationships among SRL indicators, engagement metrics, and performance outcomes rather than as definitive predictors of student success.

From a theoretical perspective, this study supports a multidimensional interpretation of self-regulated learning that integrates cognitive, behavioural, social, and emotional components within technology-enhanced learning environments. While classical SRL frameworks acknowledge motivational and affective elements, empirical studies in flipped learning contexts have often emphasised cognitive strategy use without fully incorporating emotional and social dimensions (Panadero, 2017; Dignath et al., 2023). The present findings suggest that social–emotional intelligence and affective states should be conceptualised as supporting conditions that facilitate SRL processes rather than independent causal drivers of academic performance. Learners’ emotional awareness, interpersonal communication, and collaborative skills appear to coexist with higher engagement and reflective learning practices, reinforcing the idea that SRL is a context-dependent and socially situated process.

Additionally, the integration of learning analytics with SRL theory demonstrates how digital trace data and artefact-based indicators can operationalise theoretical constructs within authentic classroom environments. Recent research has highlighted the growing potential of learning analytics to support adaptive feedback, monitor engagement patterns, and upgrade instructional design in higher education (Heikkinen et al., 2023; Wong et al., 2025). However, these analytical approaches should complement rather than replace pedagogical theory, as meaningful interpretation of behavioural data requires grounding in established educational frameworks.

This study must be considered in the context of certain limitations. First, the restricted sample size of 96 participants confines the generalizability exclusively to the DBMS course and its related domain. Although this supports internal validity for analysing correlations among SRL, engagement, and performance within the DBMS course, it hinders the generalisability of findings beyond analogous engineering education contexts. Second, self-reported social–emotional intelligence may result in biased responses. Third, the cross-sectional study design presents challenges in establishing causal relationships across semesters. This model should be extended over time to include a range of universities in future research. Fourth, employing multimodal affective sensors, behaviour indicators and emotion detectors has the potential to improve data accuracy. A further limitation concerns the psychometric validation of the adapted SEI instrument. While content and face validity were supported through alignment with established emotional intelligence frameworks, and internal consistency reliability was confirmed, full construct validity (e.g., via factor analytic techniques) was not independently established within the present sample. Consequently, the SEI measures should be interpreted as context-specific and theoretically grounded indicators rather than fully validated psychometric instruments. Future research could further employ larger samples and conduct exploratory and confirmatory factor analyses to evaluate measurement validity and dimensional structure.

The overall findings corroborate that the implementation of structured Self-Regulated Learning (SRL) interventions in flipped classrooms enhances academic achievement while simultaneously strengthening emotional resilience and fostering greater engagement. By incorporating Machine Learning (ML)-driven analytics to investigate adaptive AI systems that tailor learning trajectories based on an individual’s interests and emotional involvement, a data-driven structure could be established for upcoming hybrid educational models, leading towards the development of emotionally intelligent, self-regulating students in higher education.