Malware Detection for Android Systems using Neural Networks

The proliferation of Android malware poses an ever-growing menace to billions of mobile users worldwide. Detection systems are updated constantly to address these threats. Nevertheless, a counteraction arises in the form of evasion attacks, where an opponent modifies malicious samples in a way that causes them to be incorrectly classified as benign. In this paper, the proposed method aimed to investigate the signs of malware on Android devices, and to develop a malware detection model for Android systems based on the Drebin and the MH-100K datasets. We used each of the LSTM, MLP, and RNNs for reducing and detecting the threats and malware to enhance security over the Android systems. Each algorithm works separately and calls the sub-algorithms in the feature selection (PCA, and CFS). We used several scenarios for testing the performance of each algorithm according to the number of attributes in both datasets and the number of epochs for each algorithm. The experiment results showed the preference of results for the MH-100K dataset compared to the Drebin dataset. On the other hand, the results showed that the accuracy for the LSTM algorithm reached (98.31%) and outperformed both the MLP and the RNN algorithms for malware detection for both datasets.