Analysis of Residents'Travel Mode Choice in Medium-sized City Based on Machine Learning
This paper aims to investigate the characteristics of travel behaviors and the influencing factors on travel mode choice in a medium-sized city.Utilizing travel data from a medium-sized city in China,a random forest model embedded with a particle swarm optimization algorithm adding a variation procedure(PSO-RF)was proposed for travel mode choice prediction,due to the distinctions in prediction accuracy and modeling rationality of discrete choice model and machine learning model,as well as the characteristics and efficiency of hyperparameter optimization algorithms.The predictive accuracy,predictive mode proportion's absolute deviation,and expected simulation error were used to statistically compare the predictive performance among PSO-RF,machine learning models,and the multinomial Logit model.The SHAP(SHapley additive exPlanation)model was employed to thoroughly analyze the nonlinear relationships among individual socio-economic attributes,travel attributes,mode attributes,and residents'travel mode choices.The results indicate that PSO-RF has the highest average overall prediction accuracy(0.856),and the lowest average predictive mode proportion's absolute deviation(0.062)and average expected simulation error(0.306).Statistically significant differences in models'predictions are observed.Distance has the most prominent impact on the choice of different travel modes.The modes of walking and private cars show higher sensitivity to distance,with probability changes exceeding 35%at different distances.Individuals under 30 years old exhibit greater variations in the probability of choosing different travel modes compared to other age groups.Gender,car ownership,and bus IC card ownership notably affect the probability of choosing a bus and a private car.
urban traffictravel mode choicemachine learningmedium-sized cityparticle swarm optimizationSHAP model
NETL
NSTL
万方数据
维普
