An Integrated Hybrid Model for Predicting Urban Rail Transit Ridership Based on Network Topology and Land Use

The continuous expansion of urban rail transit networks has greatly improved system accessibility, reshaping both the magnitude and spatial distribution of ridership. To address the insufficient consideration of network characteristics in ridership prediction, this study uses Shanghai’s 2019 rail transit network as a case study and constructs a topological network graph that explicitly incorporates transfer impedance. Three centrality measures—degree, closeness, and betweenness centrality—are employed to quantify the structural roles of individual stations. Based on the spatial distribution of ridership, the K-means + + algorithm classifies stations into three types. A dual-feature prediction framework integrating network and land-use characteristics is then developed, using Decision Tree, Random Forest, and XGBoost algorithms to predict inbound ridership. Results show that models using only land-use features achieve about 70% accuracy, while including network characteristics raises accuracy to around 85%, demonstrating the importance of topological features in ridership prediction. Moreover, models optimized for different station types perform best within the clustering framework, revealing the spatial heterogeneity of ridership and the differentiated effects of network structure across hierarchical station types.