Urban Road Passenger Interpretation Based on MLP and SHAP

Abstract

Road passenger transport is China's most important mode of passenger transport. Through research and analysis of people's willingness to travel by mode, and taking into account the current social structure of China and the nonlinear and stochastic characteristics of road passenger transport, a fully-connected neural network model was built using deep learning methods to forecast the passenger transport volume of over 260 cities above the municipal level in China. A system of forecasting indicators was constructed using data on road passenger traffic and related factors for all cities in 34 provinces in China. The system is based on three main aspects: socio-demographic economy, urban transport construction, and urban fiscal policy. Finally, the SHAP model was used to calculate the Shap Values of each factor to determine the degree of influence of each factor on the dependent variable and further improve the prediction accuracy. Comparing the predicted values with the true values, the R2 of the model fit is above 60%. The accurate prediction results validate the good application of the fully connected neural network model for urban road passenger transport.