Abstract
The explosion in growth of Android malware has driven the need for accurate, efficient and interpretable Android malware detection frameworks. This paper introduces EvoMal-Net, an evolutionary artificial intelligence-driven Android malware detection system that combines the genetic algorithm (GA)-based optimization with hybrid ensemble learning. The proposed framework uses a two-stage GA process where optimal feature subsets are determined in the first stage followed by optimal ensemble voting weight optimization in the second stage to get maximum classification performance. EvoMal-Net integrates Light Gradient Boosting Machine (LightGBM) and Deep Neural Network (DNN) classifiers and adds a SHapley Additive exPlanations (SHAP) based explainability layer to the model for improving model transparency. Features extracted from static, dynamic, and contextual data sources are fused to increase the robustness of the detection. Experimental evaluation using a synthetically constructed dataset of Android malware shows that EvoMal-Net achieves 96% accuracy with a 0.95 F1-score and a 0.97 AUC while having a low false positive rate and outperforms conventional baseline models. The results confirm the effectiveness of evolutionary optimization and hybrid learning in enhancing the detection of Android malware with enhanced interpretability.