Detection of Electronic Wallets Malware Using a Hybrid Deep Learning Framework

Electronic wallet applications have become prime targets for sophisticated cyberattacks. This research proposes a hybrid deep learning framework integrating an Autoencoder, Bidirectional LSTM, Multi-Head Attention, and classification layers for Android malware detection. Evaluated on the CICMalDroid2020 benchmark dataset (11,422 samples across five malware categories), the framework achieved 99.94% accuracy in binary classification with 0.06% false positive and false negative rates, and 99.85% accuracy in multi-class classification with only 5 misclassifications out of 3,427 test samples. Ablation studies confirmed the necessity of all components. The framework outperformed 11 recent benchmark studies by +0.05% to +10.94% while reducing error rates significantly. This work represents the first specialized framework for e-wallet malware detection with comprehensive multi-class categorization and demonstrates superior performance over existing approaches.