Classifying Malware in Android Applications Using Recurrent Neural Networks and Transfer Learning Techniques

Today, malware activities are a significant security threat to Android applications. These risks are capable of stealing important information and creating havoc in the economy and our social and financial fabric. Cybercrime traffic focuses on Android apps since these continually connect to the Internet [1]. In this paper, a new technique for detecting the presence of malware in Android applications is presented, using Recurrent Neural Network (RNN) and transfer learning. The growth in the number of Android malware threats calls for a reliable and efficient method of detection of the malware. Our approach exploits the characteristics of RNNs for analysing sequential data and transfer learning for using a pre-trained model for a different task. The experimental results show that our model has high accuracy in classifying malware, verifying the reasonability of applying RNNs and transfer learning in the field of malware identification. The significance of using the graph-based approach to identify malicious parts of Android applications from the packet capture is evidenced by the experimental results, proposing that this technique performs better than traditional machine learning algorithms and can improve the current malware detection system in Android applications.