Dynamic Android Malware Detection Using Temporal Convolutional Networks
Abdurraheem Joomye, Mee Hong Ling, Kok‐Lim Alvin Yau · 2023
As new malware emerges and continues to grow in number with increasing complexity, user applications are foreseen to face greater security challenges. Hence, there is a need to leverage machine learning and deep learning (DL) approaches in Android malware detection. Temporal convolutional network (TCN), is a DL approach which uses causal and dilated 1-dimensional convolution layers to process sequential data, has not been implemented with features extracted through dynamic analysis (e.g. system calls) of Android malware applications. TCN has been shown to achieve higher accuracy in applications dealing with sequential data, with lower delay due to its parallelism and less computationally exhaustiveness. This paper implements TCN for dynamic Android malware detection. Compared to long short-term memory (LSTM), TCN is shown to achieve a similar validation accuracy and a higher steady convergence, with TCN achieving an accuracy of 89.25% and LSTM achieving an accuracy of 88.42% with 1000 system calls. Therefore, it is an effective model for Android malware detection.