JMIR Publications

ABSTRACT

The incidence of chronic diseases associated with physical inactivity is on the rise. To address this public health issue we developed a mobile app that utilizes contextual multi-armed bandits, a type of reinforcement learning algorithm, to develop personalized workout plans, adjusting task difficulty. To test the role of tailoring and choice, we develop an adaptive intervention to measure the effectiveness of personalized goal recommendation (varying task difficulty) based on online reinforcement learning. Participants are divided into three groups: user choice (no recommendation), user choice with automated recommendations (contextual bandits), and automated plans without choice (contextual bandit optimal action). The main objectives are (1) to determine the effectiveness of contextual bandits for automated goal setting, (2) to understand the role of user characteristics impacting ideal workout schedules, and (3) to explore the influence of user autonomy on recommendation effectiveness. This research will contribute to the understanding of user choice versus data-driven recommendations in mobile health interventions, potentially informing the development of more effective behavior-change apps.