Research On The Impact Of Multi-scale Built Environment On Car Ownership Behavior

With the expansion of urban scale,residents’ dependence on cars has gradually deepened,and the ensuing environmental pollution,traffic congestion,and waste of resources have become the main problems facing China.Exploring residents’ car ownership behavior and analyzing the threshold effect of different scales of built environment on residents’ car ownership can provide guidance for urban planning and traffic management.First of all,based on the existing literature research,under the premise of combining China’s traffic status and urban construction,the influencing factors of residents ’ car ownership behavior are selected.Based on the comprehensive report of the dynamic survey of labor force organized by the Social Science Survey Center of Sun Yat-sen University in 2014,the relevant data are extracted.Combined with the national and regional statistical yearbooks,the data are cleaned and processed.The individual,family,community and urban information are matched by the unified number of the report,and the data set based on individual residents is obtained.The data set is matched with the urban layer to obtain the sample data distribution map.Secondly,for the data set of residents’ car ownership behavior,considering the integrity of the data and the correlation and collinearity between the attribute variables,14 factors including individual family attributes,community built environment attributes and urban built environment attributes are selected as independent variables.Whether residents have cars as a dependent variable,analyze the correlation between their respective variables and residents’ car ownership behavior,and statistically analyze the data.Then,the Logit model,SVM,random forest model and GBDT model are established,and the accuracy rate(Acc),AUC and F1 are selected as the indexes of the evaluation model.Considering the unbalanced problem of the data set in this paper,the SMOTEENN algorithm is used to deal with the unbalanced data set of residents’ car ownership behavior.Then the results of GBDT model are compared with those of Logit model,SVM and random forest model.The results show that the indexes of GBDT model are better than those of Logit model,SVM and random forest model.Finally,based on the GBDT model with better prediction and evaluation indicators,the relationship between each attribute variable and residents’ car ownership behavior is analyzed,and the influence of selected personal family attributes and built environment attributes of different scales on residents’ car ownership behavior is quantified.At the same time,based on the GBDT model,the nonlinear relationship between the individual family attributes of the individual scale and the built environment characteristics of the community scale and the urban scale and the residents’ car ownership behavior is visualized,and the marginal effects of their respective variables on the car ownership behavior are analyzed and studied.The results show that the individual family attributes of residents have the greatest cumulative effect on car ownership behavior,which is 46.3%.The built environment characteristics of the community scale and the urban scale also have a significant impact on the car ownership behavior of residents,with an impact degree of 33.94% and 19.76%,respectively.Therefore,adjusting urban road construction,rationally planning public transportation facilities and commercial area distribution are effective means to reduce residents’ car ownership and use behavior.The research results can provide theoretical support for refined urban planning and transportation policy formulation.