Extracting Trips from Multi-Sourced Data for Mobility Pattern Analysis: An App-Based Data Example.

Area of Science:

• Transportation Science
• Data Science
• Human Mobility Analysis

Background:

• Passively-generated data (e.g., GPS, cellular) offer opportunities for human mobility analysis but require trip extraction.
• Existing methods primarily use single-sourced data (e.g., GPS, cellular triangulation), lacking approaches for multi-sourced data.
• Multi-sourced data, from technologies like GPS, cellular, and WiFi, present challenges due to temporal and spatial variances.

Purpose of the Study:

• To propose a novel framework for extracting trips from multi-sourced passively-generated mobility data.
• To address the limitations of existing methods in handling data from diverse positioning technologies.
• To enable more comprehensive human mobility analysis using increasingly common multi-sourced datasets.

Main Methods:

• Development of the "Divide, Conquer and Integrate" (DCI) framework for trip extraction.
• Application and evaluation of the DCI framework on multi-sourced app-based mobility data.
• Comparison of the DCI framework against a state-of-the-art Support Vector Machine (SVM) model for GPS data.

Main Results:

• The DCI framework successfully extracts trips from multi-sourced data with high variances in accuracy and observation intervals.
• The proposed framework significantly outperforms the SVM model designed for single-sourced GPS data.
• Obtained mobility patterns from the DCI framework align with those from external household travel survey data.

Conclusions:

• The DCI framework is an effective solution for trip extraction from challenging multi-sourced mobility data.
• This approach enhances the utility of passively-generated data for transportation applications and human mobility studies.
• The framework demonstrates robustness and accuracy, paving the way for advanced mobility analysis.