A Hybrid SMOTE-PSO-LSTM-GRU Model for Enhanced Android Malware Detection

As the use of Android devices increases, malware threats are becoming increasingly critical and often undetected by conventional methods due to data imbalance and dynamic behavior in network traffic and application activities. This study aims to answer the question of whether a hybrid deep learning model equipped with optimization and data balancing techniques can significantly improve the performance of malware detection. We propose a novel architecture that integrates SMOTE to balance the class distribution by oversampling minority malware samples, an LSTM-GRU network to learn sequential behavioral patterns, and Particle Swarm Optimization (PSO) to optimize model hyperparameters. The model is trained using a real-world dataset that includes labeled network and application activity logs. Compared with baseline models such as standard LSTM and GRU, our approach shows significant performance improvements, with an F1 score of 98.3%, an accuracy of 98.8%, a precision of 98.1%, and a recall of 98.5%. These results indicate that the proposed model not only addresses the major challenges in Android malware detection but also has strong potential for application in real-world mobile security systems.