Cluster and factor analysis on data of fatal traffic crashes in China

Abstract

In order to avoid the frequent severe crashes, it is crucial to find out the main influential factors of these crashes, according to previous crash data. The paper analyzed the importance of various factors based on the human-vehicle-road system, using two methods of multivariate statistical analysis (cluster analysis and factor analysis). The data of 166 severe traffic crashes, each causing 10 deaths and above (hereinafter referred to as severe traffic crashes), in China from 2008 to 2014 was collected. Descriptive statistical analysis of the collected data was performed, from which we extracted 23 main factors, involving human, vehicle, and road. Cluster analysis was used to classify the 23 factors. The similarities between different variables were calculated by Ward's method. With hierarchical clustering of variables, this paper formed a detailed classification system and drew the tree diagram according to the clustering process. Next, taking relevant experience into account, the importance classification results of the factors were obtained. On the basis of cluster analysis, this paper used factor analysis method to extract the common factors from the variables. We calculated the factor loading matrix by principal component analysis, rotated the matrix by maximum deviation method, and then calculated the factor scores by regression method. Finally, this paper figured out the weight of every influential factor through the variance contribution rate of each common factor and the factor score matrix, from which the 23 factors were ranked. The comprehensive analysis shows that, among all affecting factors, the main factors and their weights as follows: speeding (8.0%), ramp (6.3%), weather (6.1%), road surface (5.9%), driving experience (5.6%), road alignment (5.4%), etc. Based on this result, the following suggestions can be put forward to improve the traffic safety situation: To curb speeding and other illegal driving behaviors; To improve protection facilities at curves and ramps; To maintain the road surface condition in adverse weather; To strengthen the management and training of novice drivers; To enhance the traffic supervision during holidays. The statistical analysis methods and models used in this paper are expected to play a role in the application of big data in traffic in future.