Automated Extraction of Community Mobility Measures from GPS Stream Data Using Temporal DBSCAN Challenges with Using GPS Data

Abstract

Inferring community mobility of patients from GPS data has received much attention in health research. Developing robust mobility (or physical activity) monitoring systems relies on the automated algorithm that classifies GPS track points into events (such as stops where activities are conducted, and routes taken) accurately. This paper describes the method that automatically extracts community mobility measures from GPS track data. The method uses temporal DBSCAN in classifying track points, and temporal filtering in removing noises (any misclassified track points). The result shows that the proposed method classifies track points with 88% accuracy. The percent of misclassified track points decreased significantly with our method (1.9%) over trip/stop detection based on attribute threshold values (10.58%).