Trajectory-based clustering for enhanced attractive region mining in urban taxi services

Abstract

Trajectory data, increasingly available due to location tracking technologies, holds immense potential for intelligent traffic management and urban planning. Traditional 'attractive region' mining methods often rely on density-based clustering, neglecting the inherent path information within trajectories. To address this, we propose a novel graph-based approach for attractive region discovery. By transforming trajectory data into graphs, we effectively leverage path and connectivity information for clustering with locality-sensitive hashing. Our study introduces the pARM, pgARM, and hgARM algorithms, demonstrating their superiority over GridDBScan through experiments on real-world datasets. We employ Davies–Bouldin metric and visualization techniques to highlight the robustness of our approach, especially for datasets with varied degree distributions. Although our method may have slightly longer processing times for smaller grid sizes, it achieves execution times comparable to GridDBScan for larger grids. We rigorously analyze performance variations within our algorithms using execution time, clustering coefficient, and modularity scores, providing guidance for their optimal application.

Keywords

• Intelligent transportation
• urban planning
• trajectory data mining
• attractive region mining
• locality-sensitive hashing
• clustering