Fuzzy Clustering Based Traffic Pattern Identification

Automatic anomaly detection is of great importance in the big data era because the large volume of raw data can be accessed easily and the automatic method to analyze the data is desirable. This paper uses a framework based on fuzzy c-means clustering to detect anomaly in temporal traffic data. In this framework the sliding window is employed first to generate a collection of segments or subsequences of the time series. Then the fuzzy clustering is applied on those segments to reveal the outliers or abnormal segments in the series. The abnormal score for each segment is calculated according to the clustering results. To obtain the best setting of parameters and more meaningful abnormal scores, we design one novel performance index. The proposed approach is tested on the temporal traffic data set collected from Beijing, China, and the results demonstrate that the proposed approach can identify many valuable traffic patterns in the data.