Explainable AI-Driven Hybrid Feature Fusion for Robust Android Malware Detection

Abstract Android malware is growing rapidly, creating an urgent need for powerful and interpretable detection methods. Mobile security needs better ways to find and understand threats. The goal of this study is to create a strong, clear framework for finding Android malware. In this paper we have presented a hybrid feature fusion methodology that integrates static metadata (permissions, intents), dynamic API requests (exceeding 23,000 behavioral indicators), and DEX image-based features derived from a convolutional neural network. The approach uses a dataset of 74,268 APK samples from a public repository. Of these, 18,440 are benign and 55,828 are malicious. The dataset is split into 70% for training, 10% for validation, and 20% for testing. The experimental results show that we got an optimal outcome of 100% accuracy on the test set, suggesting a strong effectiveness of the hybrid approach. The use of explainable AI approaches helps to find the most important factors that affect classification, like questionable permissions and API requests related to the network. These techniques make it easier for nontechnical users to understand without the need to know the underlying complexity. This work is new because it combines multi-modal characteristics with interpretable AI to create a scalable, clear solution for finding Android malware in the real world.