A Novel Unsupervised Malware Detection Method based on Adversarial Auto-encoder and Deep Clustering
Lanping Zhang, Jie Yin, Jinhui Ning, Yu Wang, Bamidele Adebisi, Jie Yang · 2022 9th International Conference on Dependable Systems and Their Applications (DSA) · 2022
Malware detection (MD) is considered as one of the key techniques to solve the problem of network intrusion in the field of cyber security. Exiting MD methods are mostly based on supervised learning and require a large number of labeled datasets for training. However, labeled datasets are hard to obtain in many practical application scenarios. To solve this problem, this paper proposes an unsupervised malware detection (UMD) method based on adversarial auto-encoder (AAE) and deep clustering (DC). This method uses a large number of unlabeled data collected from internet traffic and combines a number of machine learning methods. Compared with traditional MD methods, the new method can greatly improve classification accuracy and overall system performance. This method compresses the dimensionality reduction data suitable for clustering space using AAE and DC. Experimental results are provided to show that this method can improve the MD accuracy to 87.7% under unsupervised conditions.