Blockchain-Federated Learning based Architecture for Android Malware Detection
Monalisa Maity, Priya Chaudhary, Ishita Gupta, Anshul Arora · 2025
As the number of Android applications continues to grow, the threat of malware entering Android users’ devices is also increasing. Thus, Android malware detection techniques are essential today. We propose to leverage a deep learning model consisting of an encoder and LSTM layers with enhanced scalability and privacy of federated learning to employ a secure architecture further enhanced with Blockchain for Android malware detection. We have compared the performance of our proposed architecture for 7 combinations of Android application features: Permissions, Intents, Hardware Components, Permissions and Hardware Components, Permissions and Intents, Hardware Components and Intents, Permissions with Intents and Hardware Components using metrics such as accuracy, F1 score, True Positive Rate (TPR) and False Positive Rate (FPR). For the combined feature experiments, we have used PCA for feature reduction to get better performance. Our model performs best when trained on permission features by achieving an accuracy of 96.63 %, F1 score of 96.59%, 94.85 % TPR and 1.57 % FPR.