What Are Good Situations for Running? A Machine Learning Study Using Mobile and Geographical Data

Abstract

Running is a popular form of physical activity. Personal, social, and environmental determinants influence the engagement of the individual. To get insight in the relation between running behavior and external situations for different types of users, we carried out an extensive data mining study on large-scale datasets. We combined 4 years of historical running data (collected by a mobile exercise application from over 10K participants) with weather, topographical and demographical datasets. We introduce weighted frequent item mining for the analysis of the data. In this way, we capture temporal and environmental situations that frequently associate with different running performances. The results show that specific temporal and environmental situations (hour in a day, day in a week, temperature, distance to residential areas, and population density) influence the running performance of users more than other situational features. Hierarchical agglomerative clustering on the running data is used to split runners in two clusters (with sustained and less sustained running behavior). We compared the two groups of runners and found that runners with less sustained behavior are more sensitive to the environmental situations (especially several weather and location related features, such as temperature, weather type, distance to the nearest park) than regular runners. Further analysis focused on the situational features for the less sustained runners. Results show that specific feature values correspond to a better or worse running distance. Not only the influence of individual features was examined but also the interplay between features. Our findings provide important empirical evidence that the role of external situations in the running behavior of individuals can be derived from analysis of the combined historical datasets. This opens up a large potential to take those situations specifically into consideration when supporting individuals which show less sustained behavior.

Reinforcement Learning to Send Reminders at Right Moments in Smartphone Exercise Application: A Feasibility Study

“…Following this procedure, we first clustered users from a large-scale historical running data and identified the target users [ 25 ]. The used running data contains around 440,000 runs over 4 years performed by more than 10,000 users.…”

The Design and Development of a Personalized Leisure Time Physical Activity Application Based on Behavior Change Theories, End-User Perceptions, and Principles From Empirical Data Mining

“…We combined this data set with geographical data sets to study the effect of environment variables. The results of this study showed that the most important variables were day of the week, time of day, and weather (63). Therefore, the optimization of the timing of the message was done depending on these contextual variables.…”

Optimizing Adaptive Notifications in Mobile Health Interventions Systems: Reinforcement Learning from a Data-driven Behavioral Simulator

“…First, a running dataset was used to derive the distribution P(C t |A t ) , measuring the relation between user context and running behavior. The data contains around 406K runs contributed by over 10K Dutch users while using a mobile fitness app from 2013-03 to 2017-03 [33].…”