Abstract: Network traffic is a basic data source of anomaly detection, and the accurate description of its behavioral characteristics plays an important role in real-time network abnormal behavior detection. To solve the problem of traffic anomaly detection, a logistic regression model-based network traffic anomaly detection method is proposed in this paper. By analyzing several basic characteristics of network traffic such as source IP and destination IP, the training machine of network abnormal and normal behaviors is constructed. Then, the mining model of anomaly network traffic is established by using logical regression. To valid the effectiveness of the proposed mining model, real network traffic collected by our lab is applied to test the model. Experimental results show that the proposed mining model of the network abnormal traffic is able to yield high accuracy, and achieve real-time performance as well.

Key words: logistic regression, machine learning, anomaly detection, big data analysis