A Novel Android Malware Detection Approach Based on Convolutional Neural Network

With the explosive growth of Android malware, there is a pressure for us to improve the performance of existing malware detection approaches. In this paper, we proposed DeepClassifyDroid, a novel android malware detection system based on deep learning. DeepClassifyDroid takes a three-step approach: feature extraction, feature embedding and deep learning based detection. The first and second steps perform a broad static analysis and generate five different feature sets. The last step performs malware detection based on convolutional neural networks. We evaluated our approach with different feature sets and compared with a variety of machine learning based approaches. Study shows that DeepClassifyDroid outperforms most existing machine learning based approaches and detects 97.4% of the malware with few false alarms. Moreover, our approach is 10 times faster than Linear-SVM and 80 times faster than kNN.