Feature Selection-Based Machine Learning Approaches for Detecting Android Malware with Explainable AI

With the involvement of technology, its negative sides are showing out drastically. As Android is an open-source platform that has nice customization, control and accessibility features cost-effectively, it attracts middle-class device users. As of 2023, Android has a 70.29% global market share. So, its insecurity has a tremendous effect on the mobile industry. This study represents how to efficiently detect Android malware in machine learning with various feature selection techniques like Correlation, Mutual Information, Chi-Square test and Recursive Feature Elimination to serve the optimization of the huge amount of information collected from different apps. Besides, these criteria are tested with different under-sampling and oversampling techniques. This research evaluates some more widely used algorithms: Decision Tree, Random Forest, and XG Boost. Among them, Random Forest surpasses all other algorithms. The output is evaluated with another dataset. Also, Explainable AI is used for the transparency of the models. This implies that Smote NC-based Mutual Information gives the highest outcome(98.5%) with an F1 score of 98% using a 10-fold cross-validation (CV) in Android malware detection.