Activity detection with google maps location history data: Factors affecting joint activity detection probability and its potential application on real social networks. (January 2023)

• Record Type: Journal Article
• Authors: Parady, Giancarlos; Suzuki, Keita; Oyama, Yuki; Chikaraishi, Makoto
• Is Part Of: Travel behaviour and society. Volume 30(2023)
• Page Start: 344
• Page End: 357
• Publication Date: 2023-01
• Subjects: Google maps location history -- Social networks -- Travel behavior -- Joint activities -- Passive survey methods; Transportation -- Periodicals; Population geography -- Periodicals; 303.48305
• DOI: 10.1016/j.tbs.2022.10.010
• Languages: English
• ISSNs: 2214-367X

Abstract

Highlights: We evaluated the effectiveness of using Google Maps Location History data to detect joint activities against ground truth data under experimental conditions. For Android devices, detection rates for 4-person group activities ranged from 22% under the strictest spatiotemporal accuracy criteria to 60% under less strict yet still operational criteria. Compared to Android devices, detection on iPhones markedly deteriorates. Floor area ratio (FAR) at location, activity duration, Android device ratio, open space locations and group size have non-trivial effects on joint activity detection probability. Abstract: Joint activities, despite their importance, remain poorly explained in travel behavior analysis due to lack of empirical data. This study, as an alternative to traditional travel behavior surveys (i) estimates joint activity detection rates using Google Maps Location History data under experimental conditions, (ii) quantifies the effect magnitude of factors affecting detection probability, and (iii) discusses its potential application to detect joint activities in real social networks. To do so, an experiment was conducted where participants were asked to execute daily schedules designed to simulate daily travel incorporating joint activities. For Android devices, detection rates for 4-person group activities ranged from 22% under the strictest spatiotemporal accuracy criteria to 60% under less strict yet still operational criteria. The performance of iPhones was markedly worse than Android devices, irrespective of accuracy criteria. In addition, logit models were estimated to evaluate factors affecting activity detection given different spatiotemporal accuracy thresholds. In terms of effect magnitudes, non-trivial effects on activity detection probability were found for floor area ratio (FAR) at location, activity duration, Android device ratio, device model ratio, whether the destination was an open space or not, and group size. Although current activity detection rates are not ideal, these levels must be weighed against the potential of observing travel behavior over long periods of time, and using Google Maps Location History data in conjunction with other data-gathering methodologies to compensate for some of its limitations.