Prediction And Its Characteristic Analysis Of Road Traffic Accident Under The Influence Of Environmental Factors

In recent years,with the development of the economy and the improvement of the living standards of the residents,the number of motor vehicles and drivers in China is also increasing,and road transportation has become one of the main modes of traffic for people.The level of road traffic safety is mainly determined by pedestrians and drivers,vehicles and the environment,among which environmental factors are the most complex and changeable.While road traffic brings convenience to people’s travel,the changeable driving environment also brings some losses to people’s lives and properties.Therefore,under the influence of environmental factors,the prediction of road traffic accidents has become an important research topic in the field of road traffic safety and one of the important ways to improve the level of road traffic safety.It has received extensive attention from scholars at home and abroad in the field of road traffic safety.However,existing road traffic datasets have class imbalance and low feature correlation,which will affect the prediction accuracy and generalization ability of road traffic accident prediction models.In addition,it is also very important to analyze the impact of various influencing factors on road traffic accidents,which is conducive to reducing the fatality rate of road traffic accidents and improving the safety level of road traffic travel.However,the existing road traffic accident prediction model is almost a black box model,which has the problem of poor model interpretability and cannot intuitively discover the factors environmental and other factors affecting road traffic safety.Aiming at the problems of low prediction accuracy,poor model interpretability and low model generalization ability of current road traffic accident prediction models,the main research contents of this paper are as follows:(1)In this paper,we first deal with the problem of class imbalance in the road traffic accident dataset based on SMOTEENN.After processing,we obtain a dataset with a relatively balanced number of positive and negative samples,followed by feature selection based on Pearson correlation coefficient analysis of the correlation between features and labels.Then choose SHAP and XGBoost to build a road traffic accident prediction model SHAP-XGB,predict road traffic accidents based on XGBoost,and analyze the interpretability of the overall prediction model,single feature and single sample based on SHAP.Thus,the environmental factors affecting road traffic safety and other factors related to driving safety can be clearly obtained.(2)The traditional road traffic accident prediction model is mainly constructed based on a single model,which has the advantages of simple construction and quick debugging.However,due to its own characteristics,the single model is often prone to the problem of the low generalization ability of the model on more complex datasets.In response to this problem,this paper performs model fusion based on Voting and Stacking,using multiple "good but different" fusions to obtain a fusion model with higher prediction accuracy and stronger model generalization ability.In the simulation experiment,this paper selects KNN,Random Forest,XGBoost,Light GBM and Ada Boost to build the road traffic accident prediction model respectively,and then selects the models with higher prediction accuracy for model fusion,Random Forest,XGBoost,Light GBM and Ada Boost,and obtains the fusion model,RXL-Voting,RXLA-Voting and XLAR-Stacking.The experimental results show that the fusion model proposed in this paper has a certain improvement in both AUC and F1 evaluation indexes compared with the road traffic accident prediction model based on a single algorithm.