Explainable Machine Learning Model for Android Malware Detection

This study addresses essential cybersecurity challenges in malware detection for applications by developing an explainable machine learning framework. The Stacking Ensemble approach achieves 99.89% accuracy in malware detection while maintaining high explainability through explainable AI (XAI) techniques. The research supports Vector Machines, K-Nearest Neighbours, Logistic Regression, Decision Trees, and Random Forest classifiers with ensemble strategies (Stacking and Voting), used by SHAP and LIME for transparency. The methodology shows that permissions, different API calls, and opcode-related attributes are the features to differentiate malicious applications. Experimental results show that the Stacking Ensemble, which combines individual classifiers across all metrics (accuracy, precision, recall, F1-score), offers a transparent solution for application security that addresses the black-box nature of traditional machine learning models.