AUTHOR=Murhej Mahmoud , Nallasivan G. 

TITLE=Multimodal framework for phishing attack detection and mitigation through behavior analysis using EM-BERT and SPCA-BASED EAI-SC-LSTM

JOURNAL=Frontiers in Communications and Networks

VOLUME=Volume 6 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/communications-and-networks/articles/10.3389/frcmn.2025.1587654

DOI=10.3389/frcmn.2025.1587654

ISSN=2673-530X

ABSTRACT=IntroductionThe rapid growth of advanced networking causes a significant increase in malicious threats to website data for accessing user information via phishing attacks. For the detection of phishing attacks, many works are developed based on a single data source. But, detecting the phishing attacks of different web sources was not concentrated in any of the existing works. Thus, multiple data sources, including SMS, E-Mail, and URL links, are used in this paper to detect and mitigate phishing attacks.MethodsInitially, the input data is collected from the SMS, E-Mail, and URL datasets. The contents and URLs are extracted from the datasets. Next, the textual analysis, including behavioral analysis and structural analysis, is carried out on the extracted URL. Moreover, by utilizing the Entropy Macqueen-based Bidirectional Encoder Representations from Transformers (EM-BERT) algorithm, the contents extracted from SMS and E-Mail datasets and the textually analyzed characters of the URL are transformed into vector form. Simultaneously, the CSS files and images are obtained from the URL dataset. Then, by utilizing Spherical Principal Component Analysis (SPCA), the features are extracted. Further, the optimal features are chosen by using the Cauchy distribution-based Seagull Optimization Algorithm (CSOA). Next, the phishing attack is detected using the Explainable AI SERF CoLU Long Short Term Memory (EAI-SC-LSTM) model. The recognized phishing data and URL are updated to the Blacklist; hence, any new URL, which is already on Blacklist, is reported to the user.ResultsAs per the experimental outcomes, the proposed EAI-SC-LSTM attains accuracies of 99.627% for SSC, 99.645% for PEC, and 99.541% for WPD in phishing attack detection, which are higher than the existing works. Moreover, the proposed technique detects the phishing attack within a training time of 24417 ms (PEC Dataset).DiscussionThus, cybersecurity is improved against the evolving phishing threats.