Abstract
BackgroundHealth studies and mHealth applications are increasingly resorting to tracking technologies such as Global Positioning Systems (GPS) to study the relation between mobility, exposures, and health. GPS tracking generates large sets of geographic data that need to be transformed to be useful for health research. This paper proposes a method to test the performance of activity place detection algorithms, and compares the performance of a novel kernel-based algorithm with a more traditional time-distance cluster detection method.MethodsA set of 750 artificial GPS tracks containing three stops each were generated, with various levels of noise.. A total of 9,000 tracks were processed to measure the algorithms’ capacity to detect stop locations and estimate stop durations, with varying GPS noise and algorithm parameters.ResultsThe proposed kernel-based algorithm outperformed the traditional algorithm on most criteria associated to activity place detection, and offered a stronger resilience to GPS noise, managing to detect up to 92.3% of actual stops, and estimating stop duration within 5% error margins at all tested noise levels.ConclusionsCapacity to detect activity locations is an important feature in a context of increasing use of GPS devices in health and place research. While further testing with real-life tracks is recommended, testing algorithms’ performance with artificial track sets for which characteristics are controlled is useful. The proposed novel algorithm outperformed the traditional algorithm under these conditions.
Highlights
Studies on the influence of contextual effects on health are increasingly resorting to tracking technologies such as Global Positioning Systems (GPS) to monitor mobility, which opens possibilities to continuously evaluate exposures to environmental conditions
GPS tracking generates a huge amount of geographic data which is tricky to handle in its raw form, and requires extraction of activity locations, and, possibly trips between activity locations, to be useful for health and place research
Why should we focus on extracting information on activity locations? The fact is that linking exposure and behaviour at all times, as would allow continuous GPS monitoring, does not tell us much about how structure opportunities influence behaviour and health, mainly because of selective daily mobility bias [32]
Summary
Studies on the influence of contextual effects on health are increasingly resorting to tracking technologies such as Global Positioning Systems (GPS) to monitor mobility, which opens possibilities to continuously evaluate exposures to environmental conditions. The classical approach for point cluster detection looks at the temporal sequence of recorded locations and uses a set of decision rules based on distance and time to identify clusters. This class of algorithms iteratively tests observations to determine if they remain within a given roaming distance of previous ones. Ashbrook et al [10] use a two-step procedure where, first, GPS points are flagged as significant places if the time interval with the previous point is below a certain threshold and, based on a distance criteria, clustered into locations. This paper proposes a method to test the performance of activity place detection algorithms, and compares the performance of a novel kernel-based algorithm with a more traditional time-distance cluster detection method
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
