Fuzzy Clustering for Traffic Pattern Identification by Tianjun LI Thesis Supervisor: Assistant Professor Long CHEN Department of Computer and Information Science University of Macau Anomaly detection is of great importance in the big data era because the large volume of data can be accessed easily and the automatic method to analysis the data is desirable. In this thesis, we hope to build a common method to detect anomaly in time series data using fuzzy clustering, so we presents a framework based on fuzzy c-means clustering that is applied in real traffic temporal data. In this framework the sliding window is employed first to generate a collection of segments or subsequences of time series. Then the fuzzy clustering is applied on the representations of those segments to reveal the outliers or abnormal segments. After clustering, we applied some functions to score the results using the cluster result, and then we assigned the scores to each subsequence so as to represent the anomalies. In order to obtain more meaningful abnormal scores, we designed several performance indexes to get the best settings of parameters. The proposed approach is tested on a traffic flow data collected from Beijing, China, and the results demonstrate the approach can identify many useful traffic patterns in the traffic data.