The revolution and vision of explainable AI for Android malware detection and protection

Abstract

The rise and exponential growth in complexity and widespread use of Android mobile devices have resulted in corresponding detrimental consequences within the realm of cyber-attacks. The Android-based device platform is now facing significant challenges from several attack vectors, including but not limited to denial of service (DoS), botnets, phishing, social engineering, malware, and other forms of cyber threats. Among the many threats faced by users, it has been observed that instances of malware attacks against Android phones have become a frequent and regular phenomenon. In contrast to previous studies that concentrated on evaluating the detection skills of machine learning (ML) classifiers in determining the causes, our research is primarily focused on the revolution and vision of eXplainable AI (XAI) for Android malware detection and protection. The XAI that we have presented aims to investigate how machine learning-based models acquire knowledge during the training phase. Our proposed XAI main goal is to study and figure out what makes machine learning-based malware classifiers work so well in controlled lab settings that might not accurately reflect real-life situations. It has been observed that the presence of temporal sample irregularities within the training dataset leads to inflated classification performance, resulting in too optimistic F1 scores and accuracy rates of up to 96.11%, 90.24%, and 99.48% respectively.