A convolutional neural network based Android malware detection method with dynamic fine-tuning

Abstract

Android malware detection is an important foundation for guaranteeing the security of Android ecosystem. Convolutional neural network has been applied in Android malware detection. It usually requires a large amount of training samples for building an efficient model. However, the malware data collection costs much time and resources. The lack of training samples may lead to overfitting problem. In addition, the model may become ineffective when the data distribution is significantly changed. To handle these problems, this paper proposes a new malware detection method. It firstly trains a model on an initial training set using convolutional neural network. With the upcoming of more samples, the model is updated by fine-tuning the pre-trained model on the newly labeled data. So that the pre-trained model could be dynamically updated. The experiments on the real datasets show that our method can further improve the accuracy and gmean about 1.3% and 2.4% respectively on average.