基于可解释机器学习框架的公交驾驶人风险驾驶行为影响因素

Influencing Factors of Risky Driving Behaviors of Bus Drivers Based on Interpretable Machine Learning Framework

摘要:

为改善公交驾驶人风险驾驶行为，从驾驶人人口学特征、生理和心理健康以及企业安全教育层面分析风险驾驶行为影响因素。设计公交驾驶人行为与心理调查问卷，共获取10 201份有效样本；对比4类机器学习模型，并基于可解释机器学习框架解释内部影响机制。结果表明，CatBoost模型识别的平均绝对误差（MAE）和均方根误差（RMSE）为0.190 6和0.267 1，综合性能最好；心理健康的特征重要度最高，公交驾驶人的驾驶愤怒、焦虑和工作倦怠问题越严重，越容易发生风险驾驶行为；失眠和抑郁与风险驾驶行为呈非线性关系，中度及以下失眠、抑郁与风险驾驶行为呈正相关关系，而重度失眠、抑郁与风险驾驶行为呈负相关关系；虽然企业安全教育的特征重要度较小，但是定期参加各类安全教育的公交驾驶人发生风险驾驶行为的可能性更低。

Abstract:

To accurately grasp the key points of management and education on bus drivers， the influencing factors of risky driving behaviors among bus drivers were identified from the perspectives of demographic characteristics， physical and psychological health， and company safety education. A questionnaire survey on bus drivers’ behavioral and psychological health was conducted， with a total of 10 201 valid samples collected. Four types of machine learning models were compared， and the internal influence mechanism was explained based on the explainable machine learning framework. The results indicate that： CatBoost achieves a mean absolute error （MAE） of 0.190 6 and a root mean square error （RMSE） of 0.267 1， having the best overall performance. Psychological health is the most important factor. With severe degrees of the driving anger， anxiety， and job burnout， bus drivers are more likely to commit risky driving behavior. The impact of insomnia and depression on risky driving behavior shows the non-linear relationship. The moderate and lower degree of insomnia and depression are positively correlated with risky driving behaviors， while the severe insomnia and depression are negatively correlated with risky driving behaviors； although the importance of company safety education is relatively low， bus drivers who regularly participated in various types of company safety education are less prone to risky driving behaviors.