Research On Key Driver Recognition Method Based On Data-driven Approach

The conception,“key driver”,is proposed by the department of traffic management of public security for the purpose of preventing road traffic crashes,which refers to the group of drivers who have great influence on the road traffic safety.Generally speaking,key drivers include drivers with license of heavy truck,coach,chemical tanker,etc.Such group of drivers may lead to severer crashes with far more social influence.Therefore,it is essential that to taking specific management countermeasures on them.However,which group of drivers should be concentrated on,and how to recognize drivers with high crash risk are still problems.Meanwhile,there is limitation on the public security traffic management integrated system,that is though it systemizes key driver information management,which obviously improve the efficiency,it is only a static tool,lacking the ability of analyzing and recognizing key drivers.This research collected data of 17 thousand crashes from year 2015,as well as related information such as driver,automobile,and traffic violation with a total of 310 thousand,to reveal the potential relationship between key driver and its traffic management record.To recognize key driver,two main methodologies were used in this research: statistical analysis and data mining.There conclusions could be found by statistical analysis.First,crashes caused by heavy truck drivers tended to have severer outcome,while small vehicle drivers caused far more crashes,and led to more casualties;Second,In terms of vehicle usage,non-commercial vehicle caused more crashes,for its dominant ownership;Third,New drivers with a new car had somehow high crash risk,especially when they were not well educated.In addition,the statistical analysis found that the relationship between traffic violations and the occurrence of crashes was vague.Nevertheless,drivers with four types of violations had more crash risk,that is not using safety belt when driving,driving a vehicle with inadequate safety facilities,illegal parking and overspeeding.The data mining research used a model trained by SVM(Support Vector Machine)with polynomial kernel function.Its accuracy for recognizing drivers with high crash risk was 95.51% in ideal circumstance,which showed potential for the management system to implement the key driver recognition function after further optimization and embedment.Finally,based on the research of the module of correlation analysis between drivers and crashes,the design mentality for system improvement by data-driven approach was proposed,which mainly included establishing data resource pool,strengthening sustainable application of models,constructing data analysis and decision system with low use difficulty and high pertinence.This research also has limitations.Although the data used is enormous in amount,it is a very limited sample compared with the whole traffic management information system.And additionally,there are so many underlying factors may have relationship with crash,further research on association analysis is still needed.