Directed-edge-based mining of regular routes for enhanced traffic pattern recognition from travel trajectories

Area of Science:

• * Computational Science and Engineering
• * Data Science and Analytics
• * Transportation Science

Background:

• * Trajectory analysis is crucial for understanding user transportation behaviors.
• * Existing methods for route mining and transportation mode recognition have limitations in accuracy and efficiency.
• * Global Positioning System (GPS) trajectory data offers rich insights into movement patterns.

Purpose of the Study:

• * To develop a direction-aware regular route mining algorithm for GPS trajectory data.
• * To enhance the accuracy of regular travel route extraction and transportation mode recognition.
• * To provide a robust method for analyzing complex user movement patterns.

Main Methods:

• * Systematic processing of GPS trajectory data, including preprocessing and stay region detection.
• * Segmentation of abnormal trajectories and extraction of frequent directed edges via route clustering.
• * Integration of stop-rate features for distinguishing transportation modes (e.g., public transit vs. private vehicles).

Main Results:

• * The proposed algorithm significantly reduces Mean Absolute Percentage Error (MAPE) by 56%, 49%, and 32% compared to Rules-based, CNN, and DBSCAN methods, respectively.
• * Demonstrated enhanced accuracy in both regular route extraction and transportation mode recognition.
• * Validated on the Geolife dataset, covering 1.35 million kilometers from 208 users.

Conclusions:

• * The direction-aware algorithm effectively constructs users' regular travel routes and identifies transportation modes.
• * The method offers substantial improvements in accuracy over existing algorithms.
• * Provides valuable support for applications like dynamic traffic prediction and personalized route recommendations.