关键词: 交通工程；交通安全；山区高速公路；纵坡异质性；事故严重度；事故预防；梯度提升决策树

Abstract: Aiming at the problem of complex coupling mechanism and limited comprehensive performance of the model in analyzing the factors affecting the severity of accidents, a model of multifactor coupled accident severity (MFCAS) on mountainous expressways was designed. MFCAS classified the severity of accidents based on the number of casualties and property losses as constraints and constructed a multifactor coupled feature space. A gradient boosting decision tree based on significant factor identification algorithm (GBDT-FSIA) was proposed. By using local outlier factor algorithm to remove abnormal samples and then using heuristic mean clustering algorithm to partition longitudinal slope labels, the significant influencing factor set was determined, and its coupled effect was quantified. Research results show that under typical longitudinal-slope heterogeneity conditions, the road longitudinal slope, driver age, driving experience and weather are significant factors affecting the severity of accidents on mountainous expressways. During rainy weather, drivers with less than 10 years of driving experience and over 50 years of age have the highest risk of accidents, drivers with over 20 years of driving experience and aged from 30 to 50 in high longitudinal slope environments exhibit significant risk suppression effects, and the combination of rainy days and high longitudinal slopes has a significant synergistic amplification effect. Compared with traditional models, the average accuracy of severity prediction of the proposed model can reach 85.2%, the average accident recall rate increases by 3.13%, and the AUC value increases by 2.04%.

Key words: traffic engineering; traffic safety; mountainous expressway; longitudinal slope heterogeneity; accident severity; accident prevention; gradient boosting decision tree