A comprehensive exploration and interpretability of Android malware with explainable deep learning techniques
Diptimayee Sahu, Satya Narayan Tripathy · International Journal on Information Technologies and Security · 2024
This study introduces an innovative approach to tackle evolving Android malware threats using explainable artificial intelligence (XAI) methods combined with deep learning techniques. The framework enhances detection accuracy and provides interpretable insights into the model's decision-making process. The research utilizes the CICInvesAndMal2019 dataset for training with Deep Neural Network (DNN) techniques. It incorporates Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) XAI techniques to refine the model's features and understand its predictions. This framework uses permissions and intents as static features from Android apps. The proposed framework reduces the execution time, reducing model loss to an impressively lower value of 0.26, and exhibits a commendable accuracy of 97.86%.