中国人民公安大学学报（自然科学版）

精准刻画摩托车交通事故高风险驾驶人群体特征，对开展交通安全宣传教育以及摩托车驾驶人管理具有重要意义。基于摩托车驾驶情况问卷数据，构建了包含机器学习、部分依赖图以及沙普利加和解释在内的多维解释框架对摩托车驾驶人群体特征进行分析。首先，运用SMOTE-Tomek数据平衡技术对原始数据进行平衡处理，并利用相关系数矩阵验证其数据质量；其次，采用RF模型对各自变量进行特征重要性排序，结合部分依赖图筛选得到12个对模型预测结果影响显著的自变量；最后，利用沙普利加和解释对模型预测结果进行深入分析。分析结果表明：年龄小于35岁、文化程度在初中及以下、摩托车驾驶里程不足10万公里且曾违反禁摩规定的摩托车驾驶人更容易发生交通事故。

Accurately characterizing the attributes of high-risk motorcycle drivers involved in traffic accidents is of significant importance for traffic safety awareness campaigns and motorcycle driver management. Based on a motorcycle driver questionnaire survey, a multi-dimensional interpretative framework that incorporates machine learning, Partial Dependence Plots, and Shapley Additive Explanations, was constructed to analyze the characteristics of the motorcycle driver population. First, the SMOTE-Tomek data balancing technique was applied to balance the original data, and the data quality was verified using a correlation matrix. Next, the Random Forest model was employed to rank the feature importance of the independent variables, and 12 significant variables impacting model predictions were selected through analysis with Partial Dependence Plots. Finally, Shapley Additive Explanations were used for in-depth analysis of the model predictions. The final conclusions indicate that motorcycle drivers under 35 years old, with an educational level of junior high school or below, accumulated less than 100 000 kilometers of motorcycle driving mileage, and violations in motorcycle prohibition regulations before, are more susceptible to involvement in traffic accidents.