Decision Tree Based Station-Level Rail Transit Ridership Forecasting

Abstract

AbstractThis paper presents a decision-tree based model to forecast rail transit ridership at the station level according to the surrounding land-use patterns. The canonical correlation analysis (CCA) method is used to identify key land use variables by evaluating their degrees of contribution to the rail transit station demand, which can effectively reduce dimensionality and complexity of the decision tree. A full month of Smart Card data and detailed regulatory land use plan from Chongqing, China are collected for model development and validation. The proposed model offers the capability of targeting key lane use patterns and associating them with rail transit station boarding and alighting demand at a high level of accuracy. The proposed model can reveal underlying rules between rail transit station demand and land use variables, and can be used to assist in developing the Transit Oriented Development (TOD) plans to improve land use and transit operational efficiency.