An Effective Temporal Convolutional Networks-Based Method for Detecting Android Malware Using Dynamic Extracted Features

With an increase in the number and complexity of malware, traditional malware detection methods such as heuristic-based and signature-based ones have become less adequate, leaving user applications vulnerable. Therefore, it is necessary to continue proposing and investigating new methods for Android malware detection, including machine learning (ML) and deep learning (DL) based ones. A very highly effective DL model named temporal convolutional network (TCN) which makes use of causal convolution and dilation for sequential data processing, was yet to be implemented with various dynamically extracted features from Android malware applications. In several applications, TCN has proven to be more effective with sequential data, faster due to parallelism and less computationally exhaustive. This paper proposes a new and improved method using TCN for Android malware detection with dynamically extracted feature types including system calls, binder calls and composite behaviours. The proposed method achieves high effectiveness for dynamic Android malware detection with a validation F1-Score of 99.59% and has also shown higher effectiveness than other papers using the same (CICMalDroid 2020) dataset for dynamic Android malware detection.