Hybrid Android Malware Detection and Classification Using Deep Neural Networks

This paper presents a deep learning-based framework for Android malware detection that addresses critical limitations in existing methods, particularly in handling obfuscation and scalability under rapid mobile app development cycles. Unlike prior approaches, the proposed system integrates a multi-dimensional analysis of Android permissions, intents, and API calls, enabling robust feature extraction even under reverse engineering constraints. Experimental results demonstrate state-of-the-art performance, achieving 98.2% accuracy (a 7.5% improvement over DeepAMD) on a cross-dataset evaluation spanning 15 malware families and 45,000 apps. The framework's novel architecture enhances explainability by mapping detection outcomes to specific behavioral patterns while rigorous benchmarking across five public datasets (including Drebin, AndroZoo, and VirusShare) mitigates dataset bias and validates generalization. By outperforming existing techniques in accuracy, adaptability, and interpretability, this work advances the practicality of deep learning for real-world Android malware defense in evolving threat landscapes.