Using graph to detect network traffic anomaly

Comprehensive collection and accurate description of traffic information are core problems in network traffic anomaly detection. Aiming at the lack of traffic anomaly detection in analyzing multi-time series, we propose a network traffic anomaly detection method based on graph mining. Our method accurately and completely describes the relationships among multi-time series which are used in traffic anomaly detection by time-series graph; by means of the support count of the patterns, our method mines all the frequent patterns ,which is conducive to detecting many kinds of abnormal traffic effectively; through mining the relationships among all itemsets, our method introduces weight coefficients of the itemsets, which is able to solve relationship quantification issues of multi-time series in traffic anomaly detection. The simulation results show that the proposed method can effectively detect the network traffic anomaly and achieve a higher accuracy than the CWT-based (continuous wavelet transform-based) method in term of DDos attacks detection.