Start Date
10-12-2017 12:00 AM
Description
To motivate users to continue exercising, fitness applications may need to predict their users’ exercise behavior. However, users are not homogeneous in their behaviors and different groups may need customized interventions. Motivated thus, we developed a clustering approach to group users and predict the exercise behavior for different groups. Specifically, we adapted the recency, frequency and value framework from marketing to identify different types of users. Subsequently, we selected the use of various fitness application features, such as goal setting and social network services, for physical activity prediction of the groups. Our analyses were conducted with objective data from 402 Runkeeper users collected over a year. The preliminary results identify four groups of users i.e., inactive, moderately active, active and highly active. Moreover, the use of different features predicts physical activity interval for the different user groups. The initial contributions and remaining research plan are described.
Recommended Citation
Zhou, Ya; Kankanhalli, Atreyi; and Huang, Ke-wei, "Predicting Exercise Behavior in Fitness Applications: A Multi-Group Study" (2017). ICIS 2017 Proceedings. 5.
https://aisel.aisnet.org/icis2017/IT-and-Healthcare/Presentations/5
Predicting Exercise Behavior in Fitness Applications: A Multi-Group Study
To motivate users to continue exercising, fitness applications may need to predict their users’ exercise behavior. However, users are not homogeneous in their behaviors and different groups may need customized interventions. Motivated thus, we developed a clustering approach to group users and predict the exercise behavior for different groups. Specifically, we adapted the recency, frequency and value framework from marketing to identify different types of users. Subsequently, we selected the use of various fitness application features, such as goal setting and social network services, for physical activity prediction of the groups. Our analyses were conducted with objective data from 402 Runkeeper users collected over a year. The preliminary results identify four groups of users i.e., inactive, moderately active, active and highly active. Moreover, the use of different features predicts physical activity interval for the different user groups. The initial contributions and remaining research plan are described.