Analyzing Traffic Crash Severity With Combination of Information Entropy and Bayesian Network

The analysis of severity causality for traffic crash is essential for enhancing the crash rescue responding speed, thereby reducing the casualties and property losses caused by roadway crashes. This study constructs a severity causation network to explore the relationship between risk factors and crash severity by combining information entropy and Bayesian network. The impacts of different risk factors on the severity indexes are quantitatively estimated and compared by utilizing entropy weight method, and the key factors for severity prediction are determined by considering both the weight and the accessibility. Then, the severity indexes, i.e., the number of injuries, fatalities, and the amount of property damage, are predicted with the selected key factors based on Bayesian parameter learning. The verification results confirm that compared with severity prediction utilizing all the risk factors, the prediction utilizing selected key factors do not lead to obviously precision loss. Moreover, it significantly enhances the feasibility of crash severity prediction. Due to the appropriate abbreviation of risk factors, the prediction efficiency and practical operability in crash rescue responding is improved. The findings can be utilized in analyzing the severity causation of traffic crashes, which is serviceable for managers to find effective measures to improve traffic safety as well as reduce the casualties and property losses caused by traffic crashes. By providing the severity causation network, this study facilitates the prediction of severity level, which can provide valuable reference information for a crash response.