Mining Commuting Patterns of Urban Residents From Taxi Trajectory

Understanding urban spatial structure and resident commuting behaviors from taxi trajectories is critical for transportation planning and sustainable urban development. This paper presents an integrated framework for mining commuting patterns, combining spatial clustering and functional area identification. A novel algorithm, Grid Density Peak Clustering with Standard Deviation Weighted Distance and Fuzzy Weighted Natural Neighbors (GDPC‐SFNN), is proposed to address limitations in traditional density‐based clustering, such as parameter sensitivity and spatiotemporal feature integration. GDPC‐SFNN enhances clustering accuracy by dynamically weighting features via standard deviation distances, automating neighbor selection through the natural neighbor method, and adopting divide‐and‐conquer for outlier handling. A Work–Residence Index differentiates urban areas into work, residential, and neutral zones, enabling identification of four commuting patterns: high‐volume short‐distance, high‐volume long‐distance, low‐volume short‐distance, and low‐volume long‐distance. Experiments on real‐world taxi data show GDPC‐SFNN outperforms state‐of‐the‐art methods in clustering accuracy and efficiency. These findings provide a data‐driven basis for optimizing public transport and addressing job–housing imbalance, showcasing taxi trajectory analytics' utility in urban governance.