| Public transportation as an important part of urban transportation plays an important role in easing urban traffic congestion,reduce traffic pollution and improve the comprehensive management capabilities cities.At the same time,the occurrence of public traffic accidents is also unavoidable due to the complex urban road traffic environment.As road traffic participants and recipient of road traffic information as well as the decision-makers,bus drivers play a key role in the occurrence of road traffic accidents.The research on accident risk factors of bus drivers has also attracted more and more scholars’ s concern.How to dig out the key factors that affect the risk of accident bus drivers to improve bus operations unit safety management and road traffic safety plays an important role.With the advent of the information age and the era of big data,the rapid development and application of data mining technology provides a new perspective for solving problems and discovering knowledge.Based on the above background,this research uses 11,342 data involving violations and accident from a bus operating unit in a certain city in China within four years,and combines data mining technology with bus driver risk factors to explore the hidden knowledge in bus driver violations and accident data.Public road traffic accidents are the result of the joint action of human,vehicle,road and other factors.In order to effectively dig out the factors that affect the risk of bus driver accidents,It focuses on the analysis of the distribution characteristics and influence mechanism of factors such as the time of bus drivers’ violations and accidents,drivers,bus vehicles and routes.On this basis,based on the general characteristics and transaction characteristics of bus drivers,the characteristics of the risk bus driver identification model are constructed from the four dimensions of time characteristics,time series characteristics,trend characteristics and attribute characteristics,then use SQL data processing technology and the storage process realizes the extraction of data,constructs the bus driver accident risk analysis database.In order to improve the effectiveness of the model,a multi-criteria integrated feature selection method based on ensemble learning is proposed,then uses the XGBOOST integrated tree algorithm to build the model on the basis of resampling the data.Then use the Bayesian optimization method to optimize the model parameters.The results show that the AUC before and after parameter adjustment is increased by 2.79% to 96.31%,which is significantly better than the naive Bayes,GBDT and other models.In addition,the model can effectively identify the risk status of bus drivers with an accuracy of 98.66%.In order to improve the interpretability of the model,the traditional interpretation method and Shap partial interpretation method are combined with the established risk bus driver identification model,and the key factors affecting the model results are analyzed.The results show that the dynamic data such as violation characteristics,accident characteristics and management characteristics constructed by this research play an important role in predicting the risk of bus driver accidents.Based on the results of this research model,the influencing factors of bus driver accident risk are discussed from the three aspects of driver characteristics,violation accident management characteristics,and bus and routes,which provides a theoretical basis for bus operators to improve management and formulate reasonable and effective management measures. |
