AUTOMATIC TIME-ACTIVITY CLASSIFICATION BASED ON GLOBAL POSITIONING DATA

Abstract

Background and Aims: Air pollution epidemiological studies are increasingly using global positioning system (GPS) to collect time-location data. However, remarkable uncertainties in GPS data create challenges for reliably characterizing time-activity patterns from the raw GPS data. We aimed to develop models to classify people’s major time- activity patterns directly from the GPS tracking. Methods: We collected 131 person-days of GPS time-activity data in 2008 for 47 residents in Southern California. Time-activity patterns were coded for each subject based on GPS recordings, time-activity diary logs, and follow-up interviews. Using similar methods, 7-day validation data were collected from three research staff in 2010. We developed, evaluated, and compared two models for extracting time-activity patterns (i.e. indoor static, outdoor static, outdoor walking, and in-vehicle travel) from the GPS data. A rule-based model developed user-defined rules based on time, speed, and spatial relationship of the GPS points, while a logistic model trees (LMT) model required minimal user-input by using existing software that combined a tree structure with logistic regression. Results: Indoor static, outdoor static, outdoor walking and in-vehicle travel accounted for 82.7%, 6.1%, 3.2% and 7.2% of the points in the main dataset, respectively. Both the rule-based and the LMT models correctly classified more than 90% of reported indoor static points and more than 70% of reported in-vehicle travel points, with Type 1 error (over-estimation) and Type 2 error (underestimation) <9% for indoor static classification and <29% for in-vehicle travel classification. Both models performed the worst in identifying outdoor static points. The performance of the LMT model improved significantly when using training data from the 7-day validation dataset likely because of more accurate diary data. Conclusion: Our models can successfully identify static and in-vehicle travel points from raw GPS data. It is more challenging to separate outdoor static from indoor static points.