ConvDroid: Lightweight Neural Network based Andoird Malware Detection

Sifan Wu, Xi Xiao
2019 Australian Journal of Intelligent Information Processing Systems  
The explosive amount of Android malware have threatened the security of legitimate users. In Recent years, with the development of the neural network, more and more research is focusing on detecting malware based on the neural network. Where, most of these techniques are depending on the complex feature engineering process and have a resource and time expensive classification neural network. In this paper, we propose a novel lightweight convolution neural network model, ConvDroid, with the
more » ... m call sequences as features. Different from the existing n-gram models utilizing system call sequences, we construct a two-layer convolution neural network to learn the global information of sequences instead of limited windows. Furthermore, we assess ConvDroid for accuracy, efficiency using a dataset consisting of 3567 malicious applications and 3536 benign ones. Our experiments show that ConvDroid achieves an accuracy of more than 98% with a low time cost. We further demonstrate ConvDroid's superiority against state-of-the-art approaches.
dblp:journals/ajiips/WuX19 fatcat:3foe6cioibeqnptbxs35xxlqf4