To solve the problems caused by traffic congestions in urban areas，this study proposed a comprehensive method to evaluate and predict traffic states.Based on three discriminating factors of traffic flow，average speed and travel time delay，a dynamic traffic state discrimination model using the fuzzy C-means clustering algorithm was established first to determine the states of all samples.Then， the traffic situation prediction model is constructed based on the support vector machine model，and the clustering model is used to evaluate the traffic states and determine the road segment that needs to be traffic controlled.Finally，the data envelopment method was developed to identify the priority of each road section for more precise traffic control.Based on the empirical data collected in Changsha，China， this method was validated，the research results could provide substantial supports for decision makers to take proactive measures to release traffic congestions in urban areas.