Abstract: Traffic accident data may contain meaningful patterns of traffic accident, such as correlations between traffic accident and weather, time, road, etc. It is worthy of in-depth study. In general, the traffic accidents are correlated with weather, time and road. However, the correlation effect is different among various regions, which means there are local correlations between those factors and traffic accidents. It is valuable to reveal the relation between these factors and traffic accidents by analyzing local correlations. The paper presents a method to discover local correlations in traffic accidents. Firstly, the method extracts accident-prone road segments, each of which contains location, time and some other related accident information. A cluster-supported local correlation visual analysis method is presented to analyze accident-prone road segments: some histograms of these factors (weather histogram, time histogram) are used to feature accident-prone road segments, and a cluster algorithm is applied to analyze accident-prone road segments based onthe similarity of histograms. The cluster results are further interactively analyzed in linked-views to discover local correlations. The method is used to analyze traffic accident data of Hefei by specialists, and some meaningful local correlations are found, which demonstrates the method’s effectiveness.

Key words:  traffic accidents, visual analysis, local correlations, accident-prone road segments