Related Experiment Video
Updated: Jan 3, 2026
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
Published on: December 11, 2015
Prediction of human active mobility in rural areas: development and validity tests of three different approaches
Gijs Klous1,2, Mirjam E E Kretzschmar3,4, Roel A Coutinho3
1Julius Centre for Health Sciences and Primary Care, University Medical Centre Utrecht, Utrecht, The Netherlands. g.klous@umcutrecht.nl.
Background/Aim:
Active mobility may play a relevant role in the assessment of environmental exposures (e.g. traffic-related air pollution, livestock emissions), but data about actual mobility patterns are work intensive to collect, especially in large study populations, therefore estimation methods for active mobility may be relevant for exposure assessment in different types of studies. We previously collected mobility patterns in a group of 941 participants in a rural setting in the Netherlands, using week-long GPS tracking. We had information regarding personal characteristics, self-reported data regarding weekly mobility patterns and spatial characteristics. The goal of this study was to develop versatile estimates of active mobility, test their accuracy using GPS measurements and explore the implications for exposure assessment studies.
Methods:
We estimated hours/week spent on active mobility based on personal characteristics (e.g. age, sex, pre-existing conditions), self-reported data (e.g. hours spent commuting per bike) or spatial predictors such as home and work address. Estimated hours/week spent on active mobility were compared with GPS measured hours/week, using linear regression and kappa statistics.
Results:
Estimated and measured hours/week spent on active mobility had low correspondence, even the best predicting estimation method based on self-reported data, resulted in a R2 of 0.09 and Cohen's kappa of 0.07. A visual check indicated that, although predicted routes to work appeared to match GPS measured tracks, only a small proportion of active mobility was captured in this way, thus resulting in a low validity of overall predicted active mobility.
Conclusions:
We were unable to develop a method that could accurately estimate active mobility, the best performing method was based on detailed self-reported information but still resulted in low correspondence. For future studies aiming to evaluate the contribution of home-work traffic to exposure, applying spatial predictors may be appropriate. Measurements still represent the best possible tool to evaluate mobility patterns.
