关键词: 行人交通安全；CIDAS 数据；多次重复的 K 折交叉验证；随机森林模型；基于排列 的特征重要性算法；卡方检验

Abstract: Pedestrian road traffic accidents are a common type of traffic accidents. In order to build an effective pedestrian traffic safety prevention and control system, the China in-depth accident study (CIDAS) data is used for analysis and research. After using repeated rounds of K-fold cross- validation to evaluate and confirm that the random forest model has statistical power on this data, the permutation feature importance algorithm is used to quantify the features that affect pedestrian traffic accidents. The data of important features are then statistically analyzed and chi-squared test is used to determine the effect of randomness. The research shows that the number of accident participants, age group, accident occurrence time and maximum speed limit are the most important features affecting the consequences of pedestrian traffic accidents. The overall trend shows that the more people involved in the accident, the more serious the consequences of the accident; for people aged 13 and above, the older the pedestrian, the more serious the consequences of the accident; in the accidents that occurred between 0:00 to 4:00 in the morning, the severity of the accidents is significantly higher than that in other time periods; the consequences of accidents on roads with a speed limit of 80 km/h and above are more serious.

Key words: Pedestrian traffic safety; CIDAS data; Repeated rounds of K-fold cross-validation; Random forest model; Permutation feature importance; Chi-square test