The passenger volumes of metro stations are important parameters for metro planning and design. In this paper, a random forest regression model was proposed to predict the inbound/outbound passenger volume and the total passenger volume of each station during the workday, based on the Metro AFC data and POI data of Shanghai, China. In this model, the dependent variables include the inbound/outbound passenger volume and total passenger volume of each metro station, and the independent variables include the relative proportion of 15 types of POI, the number of residents around metro stations, the number of jobs around metro stations, whether it is a transfer station, the number of years of operation and the location of station. The results show that both the R-square value of the inbound volume and total volume are 0.94, and that of the outbound volume is 0.95. The average relative error of the model is about 20%, which means the prediction accuracy of the proposed model is good. Moreover, the variable importance method based on permutation was used to analyze the factors that affect the passenger volumes of stations significantly, and the top 3 variables of importance were obtained: the relative proportion of catering service POI, the relative proportion of transportation facility POI, and whether it is a transfer metro station. The results of this study are useful for metro station planning and design, and metro operation organization.

passenger volume prediction;metro station;random forest regression;AFC data;POI data