Mining the Critical Factors of Hazardous Material Road Transportation Accidents From Chinese Investigation Reports

Abstract

Rather than reducing losses after an accident, identifying critical accident factors is a better approach because it can be used to prevent accidents. Hazard materials (HAZMAT) road transportation is a complex sociotechnical system that involves three major procedures. Staff, vehicles, and tanks directly participate in this transport system, while different companies and government departments manage or supervise personnel and equipment. Moreover, all these stakeholders must obey ten types of industrial standards or regulations. The accident investigation reports contain the critical factors of the accident. Specifically, using a series of natural language processing (NLP) technologies, the accident causes in 137 Chinese reports are extracted, classified, and formatted. Then, a Bayesian network (BN) is constructed using the structural restrictions learning algorithm based on structured accident data from the reports. Finally, critical accident factors are revealed through sensitivity analysis of the BN. The results show that inadequate or delayed operation, which includes subordinate management factors and supervision factors, is the most critical factor among the five factors grouped. Actual accident data can explain the results. Combining the actual data and accident model, the accident analysis approach proposed in this paper considers the structural relations and frequency of accident factors.