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

Area of Science:

• Mobile Health (mHealth)
• Behavioral Science
• Machine Learning

Background:

• Just-in-time adaptive interventions (JITAI) are effective in mobile healthcare.
• Optimizing intervention timing is critical for user engagement.
• Reinforcement learning (RL) offers a promising approach for adaptive intervention delivery.

Purpose of the Study:

• To apply RL in a smartphone exercise app to promote physical activity.
• To determine the optimal timing for delivering adaptive notifications using user context.
• To evaluate the feasibility and user experience of an RL-driven JITAI system.

Main Methods:

• Developed an RL model to adaptively schedule notifications based on user context (time, calendar).
• Conducted a four-week trial with 7 participants using a smartphone exercise application.
• Analyzed user reaction data and collected qualitative feedback via questionnaires and interviews.

Main Results:

• 83.3% of adaptive reminders prompted user action within 50 minutes.
• 66.7% of physical activities occurred within 5 hours of a reminder.
• The RL model demonstrated usability in delivering timely exercise prompts.

Conclusions:

• Reinforcement learning is a viable technique for optimizing JITAI in mHealth.
• Adaptive notification timing can effectively promote physical activity.
• Further refinement of timing algorithms can enhance intervention effectiveness.