The metro system has been a mainstream of public transportation as it can alleviate the pressure of the urban traffic. As a precondition to design and construct a new metro station, it is crucial to understand the daily ridership in it. In this study, a combined random forest have been proposed for prediction. The combined random forest fuses random forest and geographical random forest, which can combine the advantages that random forest and geographical random forest can effectively reduce variance and bias respectively. To demonstrate the performance, three prediction accuracy indicators and Moran’s I test were employed. These models were implemented and validated on real-world metro station ridership data in Shenzhen, China. The results demonstrate the superior performance of the combined random forest based on the accuracy indicators. The reason is considered that the combined random forest accounts for the spatial heterogeneity according to the Moran’s I test.

CICTP