Analysis of accident causes and severity association rules at intersections based on an improved Apriori algorithm

The causes of accidents at road intersections are complex, and the accurate identification of high-risk factor combinations is crucial for improving traffic safety. This study aims to clarify the specific risk patterns of different intersection types, reveal their intrinsic correlations with accident severity, and provide data-driven support for safety interventions. A modified Apriori algorithm, optimized with hashing technology and enhanced by multi-level association analysis, is proposed to improve the efficiency of mining association rules from complex traffic data. Using 10,122 accident records from 2020-2022 in the Crash Report Sampling System (CRSS) from the National Highway Traffic Safety Administration (NHTSA), this study analyzes the association characteristics of 18 key variables (e.g., human, vehicle, and road factors) across six typical intersection types, adopting the Bow-tie model as a systematic analytical framework. The results indicate that conflicts between pedestrians or non-motorized vehicles constitute a universal, fundamental risk source across all intersection types, while distinct intersection types exhibit specific fatal risk chains (e.g., the unique risk of reversing accidents at Y-Intersection and right-side collisions with fixed objects at roundabouts). The theoretical contribution lies in verifying the effectiveness of the modified algorithm in processing complex traffic data and, through the Bow-tie model, advancing the analysis from simple risk identification to a deeper interpretation of accident mechanisms. It provides directly implementable countermeasures for specific issues.These research findings can provide precise support for traffic management departments in formulating accident prevention strategies and optimizing intersection safety facilities, thus offering significant theoretical and practical value.