DL-AMDet: Deep learning-based malware detector for android

Abstract

The Android operating system, with its market share leadership and open-source nature in smartphones, has become the primary target of malware. However, detecting malicious Android processes has become a significant challenge because of the complexity of size, length, and associations of various important and distinctive elements of Android applications, such as API calls and system calls. In this paper DL-AMDet, a deep learning architecture is proposed to detect Android malware applications based on its static and dynamic features. DL-AMDet consists of two main detection models the first one uses CNN-BiLSTM deep learning method for detecting malware using static analysis. The other model utilizes deep Autoencoders as an anomaly detection model to identify the malware based on dynamic analysis. The performance of the DL-AMDet architecture is evaluated using two different datasets. The results show that DL-AMDet achieves a competitive malware detection accuracy of 99.935 % for static and dynamic analysis models combined. Additionally, the results emphasize the significance of CNN-BiLSTM and Deep Autoencoders models used in DL-AMDet to outperform the existing state-of-the-art techniques.