Classification of Android apps and malware using deep neural networks
Robin Nix, Jian Zhang · 2017
Malware targeting mobile devices is a pervasive problem in modern life. The detection of malware is essentially a software classification problem based on information gathered from program analysis. We focus on classification of Android applications using system API-call sequences and investigate the effectiveness of Deep Neural Networks (DNNs) for such purpose. The ability of DNNs to learn complex and flexible features may lead to timely and effective detection of malware. We design a Convolutional Neural Network (CNN) for sequence classification and conduct a set of experiments on malware detection and categorization of software into functionality groups to test and compare our CNN with classifications by recurrent neural network (LSTM) and other n-gram based methods. Both CNN and LSTM significantly outperformed n-gram based methods. Surprisingly, the performance of our CNN is also much better than that of the LSTM, which is considered a natural choice for sequential data.