Explainable AI-Based Malware Detection for Protecting Assistive Learning Systems Used by Students with Special Needs

Abstract

The use of assistive learning technologies, adaptive educational platforms, and digital communication systems has greatly facilitated education in terms of accessibility for students with special needs. The growing use of these technologies, however, has led to a greater risk of being targeted by malware and other cyber threats that can impact the safety of sensitive education information. Sophisticated and evolving malware are often not detected by traditional detection methods, and many deep learning methods are not interpretable. In order to cope with these challenges, the present study presents an Explainable Artificial Intelligence (XAI) based malware detection approach for the security of assistive learning systems. The proposed architecture is based on both structured malware features and deep feature representations derived from convolutional neural networks and learning modules based on Transformer models. A feature refinement mechanism based on attention and a conditional feature fusion strategy are used to improve the discriminative learning and the classification of malware. For better transparency, Shapley Additive Explanations (SHAP) are included to offer interpretable insights into classification decisions. Experimental testing shows that the proposed approach is effective on both the SOMLAP and the EMBER benchmark malware datasets with accuracy (99.65%), precision (1.00), recall (0.996) and F1-score (0.998) better than any conventional machine learning approach. The proposed framework offers a safe, solid, and transparent cybersecurity solution to protect and secure assistive educational systems.