Android Malware Detection using Sequential Convolutional Neural Networks

Abstract

As the largest Operating System in the smart phone market, Android is gaining more adherents in recent years, which makes it increasingly important to detect Android malware correctly and efficiently. In this paper, we proposed a way to detect it with sequential convolutional neural networks. In the pre-processing work, Android package files were dissected into Dalvik operation codes. These codes were treated as text sequence in our designed model. This model was trained on 3000 samples, and tested on two large datasets. One with 10,174 samples where our model reached an accuracy of 90% in 1.2 seconds and the other with 20,083 samples where its accuracy was 88% and cost 2.4 seconds.