Modeling And Analyzing Influential Factors Of Urban Pedestrian Safety Based On Social Sensing Data

With the accelerated urbanization process,road safety has become an issue with widespread concern.Among all the road users,pedestrians are one of the most vulnerable groups that easily get injured or even die in traffic accidents due to lack of protection measures.Analyzing the extent to which the built environment can influence the occurrence of pedestrian crashes plays an essential role in pedestrian safety programs.With pedestrian crash data,social sensing data in Changning District of Shanghai,an analytical framework was established based on LightGBM and SHAP.The main contents are as follows:1)The network kernel density estimation was first used to generate the pedestrian collision density surface.This research employed multi-source social sensing data(points of interest,street view images,house price and pedestrian volume)for deriving street-level and neighbourhood-level characteristics of the environment.2)Four kinds of models of Random Forest,Gradient Boosting Decision Tree(GBDT),XGBoost,LightGBM are constructed by controlling the consistency of the training set,validation set and test set,showing feasibility of employing social sensing data for deriving the characteristics of the environment and carry out microscale pedestrian crash modeling.By comparing the goodness of fit,it is found that LightGBM models are more superior than others.Four models were then established with LightGBM by type of the road and time of the day.The performances of the models suggested that the variables could largely account for the variation in the distribution of pedestrian crashes.3)By performing SHAP analysis,the influence of the built environment was further explored both globally and locally.The results indicate the usefulness of social sensing data in picturing local characteristics of the road environment and the advantages of the framework in detecting and analyzing nonlinear relationships between contributory factors and pedestrian crashes.This could shed light on the decision-making for improving urban pedestrian safety.