Empowering Android Malware Detection: A Deep Learning Ensemble with Optimal Features
Mehedi Hasan Shakil, Md. Mynul Hasan · 2023
Because of the rapid expansion of mobile devices and the growing reliance on mobile applications, there has been an increase in Android malware threats, necessitating the development of sophisticated detection systems. This research proposes a technique for detecting Android malware that makes use of several deep learning models. The goal is to provide an accurate and dependable method for classifying mobile applications as malicious or benign. To assure data integrity, the Drebin dataset is preprocessed. RFECV was employed to obtain the optimal features for deep learning algorithms such as Bi-LSTM, Bi-GRU, and 1D CNN. The dataset was trained using the selected features for each deep learning algorithm based on the feature selection method. The ensemble technique, employing a weighted average method, was utilised to combine the predictions of the individual models. Extensive testing revealed that combining Bi-LSTM and CNN in the ensemble technique achieves the greatest accuracy of 98.99%. This combination highlights the efficiency of combining various models strengths for Android malware detection.