Android Malware Detection Based on Deep Learning
Jianming Zhang, Futai Zou, Junru Zhu · 2018
This paper proposes an Android malware detection method based on deep learning model. The input data is divided into two parts, the static features, and the dynamic features. In this paper, the Android code is deeply analyzed through reassembling to obtain the 485-dimensional static features. The dynamic temporal characteristics of the changes are obtained through the simulator operation, and the 23-dimensional dynamic features are achieved through the process of the RNN neural network. Our experiments show that the combination of static and dynamic characteristics is an excellent way to detect Android malware. Based on a large open Android dataset, this deep learning model can significantly improve detection performance compared with traditional shallow models such as LR and SVM. This method is quite efficient and practical to improve the security of the Android system.