Predicting clusters of physical activity based on individual characteristics: an event-based ecological momentary assessment study

Planning can help bride the physical activity intention behaviour gap, but creating plans has proven burdensome for individuals. Personalized plan recommendations can alleviate this burden and improve plan quality. For this, we need to know the types of physical activities individuals perform. This study aimed to identify clusters of physical activities and predict these clusters based on static and dynamic person variables.In a 14-day ecological momentary (EMA) assessment study, 52 participants completed diaries, and questionnaires after the performance of a physical activity of at least 5 minutes. Clusters of physical activities were identified. Clusters, activity domain and location were predicted using conditional random forest algorithms based on static and dynamic person variables. A six-cluster solution was identified. Identified clusters were ‘active transport activities’, ‘work-related activities’, ‘household activities’, ‘organised sport activities’, ‘in the city activities’ and ‘in nature activities’. We were able to predict up to 63% of clusters, exceeding a baseline comparison. Predictions were largely based on static characteristics, most notably sociodemographic information. Clusters of physical activity were identified and could be predicted, primarily by static characteristics. We discuss the added value of clusters of physical activity, implications for intervention development and suggestions for future research.