STUDY ON CLASSIFICATION METHOD FOR BIKE SHARING STATIONS CONSIDERING TEMPORAL DEMAND VARIATIONS

Abstract

This study proposes a station-level classification method that explicitly captures temporal demand variations in bike-sharing systems. Using one month of operational data from 47 stations (12,011 trips) in Nerima Ward, Tokyo, rental and return counts were aggregated into six-hour intervals for weekdays and holidays, yielding 16 temporal features. K-means clustering was applied, and the elbow method together with silhouette analysis indicated that three clusters were appropriate. The resulting groups (1) commute-oriented (5 stations), (2) baseline-use (39 stations), and (3) major hubs (3 stations)—exhibited markedly different peak periods and demand volumes. Although no geographic variables were included, population-density and rail-proximity patterns emerged naturally, suggesting that temporal features implicitly encode spatial context. The proposed demand-based clustering can enhance demand-forecasting accuracy and inform station rebalancing and incentive strategies, providing a more nuanced operational framework than conventional location labels.