Feature Mining of Shanghai Metro Commuters Based On K-Prototypes Clustering Method

Abstract

Urban rail transit passenger travel behavior exhibits pronounced spatio-temporal heterogeneity, which poses significant challenges for accurately capturing passenger flow dynamics and enabling precise short-term travel forecasting. Unlike most existing studies that focus on station-level predictions, this research innovatively explores the spatio-temporal characteristics of travel behavior from the perspective of individual passengers. Leveraging one-card transaction data from Shanghai’s rail transit system between April 1 and 30, 2015, we extract individual travel trajectory data in the form of OD chains through an OD relationship matching algorithm. A 13-dimensional clustering feature is constructed using a three-dimensional framework encompassing temporal, spatial, and travel intensity attributes. Notably, we introduce frequent travel OD share data as a critical metric to quantify the regularity of passenger travel patterns. To address the complexity of mixed-type data (including categorical and numerical variables), the K-Prototypes clustering algorithm is employed, demonstrating superior performance in handling heterogeneous datasets compared to traditional methods. The clustering results categorize passengers into four distinct groups, with the commuter category—accounting for 40% of the total—exhibiting the strongest spatial regularity. Further analysis of travel patterns across categories provides empirical evidence for identifying the primary sources of metro passenger flow, offering actionable insights for optimizing urban transit planning and demand management.