Why an Android App Is Classified as Malware

Machine learning–(ML) based approach is considered as one of the most promising techniques for Android malware detection and has achieved high accuracy by leveraging commonly used features. In practice, ML classifications only provide a binary label to mobile users app security analysts. However, stakeholders are more interested in reason why apps classified malicious both academia industry. This belongs research area interpretable but specific domain (i.e., detection). Although several methods have been exhibited explain final classification results many cutting-edge Artificial Intelligent–based fields, until now, there no study interpreting an or unveiling domain-specific challenges. this article, fill gap, we propose novel ML-based (named XMal ) classify with result meanwhile. (1) The first phase hinges multi-layer perceptron attention mechanism also pinpoints key features related result. (2) second aims at automatically producing neural language descriptions interpret core behaviors within apps. We evaluate behavior description human in-depth quantitative analysis. Moreover, further compare existing Drebin LIME) demonstrate effectiveness . find that able reveal accurately. Additionally, our experiments show can some samples misclassified classifiers. Our peeks into through