Driving Style Recognition And Influencing Factors Analysis Of Online Car-Hailing Drivers

With the development of economy and society,people’s travel needs are constantly derived.As a product of "Internet +" and the sharing economy,online car-hailing travel services meet the diverse needs of people’s travel modes,and quickly occupy the market with its advantages of convenience and comfort.However,due to the heterogeneity of the car-hailing drivers and the particularity of the car-hailing service model,their driving style is susceptible to multiple factors,and driving style is closely related to road safety.Therefore,in order to ensure the steady driving style of online car-hailing drivers and improve road traffic safety,it is very necessary to study the driving style of online car-hailing drivers.Focusing on the online car-hailing drivers,this paper studies the classification and causative mechanism of their driving style.Specific research contents are as follows:First,a natural driving real car test is carried out to obtain the vehicle kinematics data and video data inside and outside the car of the online car-hailing drivers in real road scenarios.Besides,a questionnaire survey is designed to obtain the driver’s personal information data.Data preprocessing such as data splicing,data cleaning,data smoothing,data conversion,time correction,and data merging is performed on these three kinds of data.Next,a driving maneuver-based driving style classification framework is proposed.The three driving maneuvers of turning,acceleration and deceleration are extracted from the vehicle kinematic data based on a threshold-based endpoint detection approach,the statistical feature parameters are extracted from the time series data to represent these driving maneuvers,and the principal component analysis is carried out to reduce the number of statistical parameters and derive more representative parameters.Then,we applied the k-means algorithm to cluster the respective maneuvers into different classes,and determine the number of clusters according to the clustering evaluation indicators.Finally,the attributes of different clusters are analyzed to determine the driving style corresponding to each cluster.Then,the influencing factors of driving style are extracted to construct a driving style recognition model based on the particularity of online ride-hailing drivers.According to the particularity of online car-hailing services,three influencing factors of driving style including environmental factors,operating characteristics and personal characteristics are extracted.Two representative statistical models,multinomial logit model and generalized additive logit model,and four integrated learning models,Random Forest,XGBoost,Lightgbm,and Catboost are selected to identify driving styles,and the Bayesian optimization algorithm is used to optimize the hyperparameters of the integrated learning model.Accuracy rate,precision rate,recall rate,F1-score,ROC curve,and AUC are selected as model evaluation indicators,and the optimal model is selected for subsequent research.The results show that the XGBoost model has the best performance in each evaluation indicator.Finally,the SHAP framework(Shaply Additive ex Planation)is utilized to explain the results of the optimal model and analyze how the selected factors affect the driving style of online car-hailing drivers.Based on feature importance,total effect,main effect,and interaction effect,the impact of a single feature factor on driving style and the interactive impact of multiple feature factors on driving style are analyzed for three driving styles.The changing characteristics of driving style under different driving stages,driving duration,driving distance,and road characteristics are analyzed emphatically.The results show that the selected factors have significant influence on driving style,and there are obvious interaction effects among these factors on driving style.In summary,this article explores the entire research process of driving style classification and influencing factors analysis of online car-hailing drivers.The driving style is classified based on vehicle kinematics data.Besides,statistical models and integrated learning models are utilized to recognize driving style based on the factors associated with driving style.Finally,SHAP is utilized to analyze the relationship between the factors and driving style.The results bring new ideas for driving style research,which are of great significance to the safety supervision of online car-hailing platforms and the development of driver assistance systems.