Flow Graph Anomaly Detection Based on Unsupervised Learning

Abstract

In this paper, a flow graph anomaly detection framework based on unsupervised learning is proposed. Compared with traditional anomaly detection, graph anomaly detection faces some problems. Firstly, the training of a reasonable network embedding is challenging. Secondly, the information data in the real world is often dynamically changing. Thirdly, due to the lack of sufficient training labeled data in most cases, anomaly detection models can only use unsupervised learning methods. In order to resolve these problems, three modules in the framework are proposed in this paper: preprocessor, controller, and optimizer. Additionally, a reasonable negative sampling strategy is applied to generate negative samples to deal with the lack of labeled data. Finally, experiments on real-world data sets are conducted, and the experimental results show that the accuracy of the proposed method reaches 87.6%.