An interpretable health behavioral intervention policy for mobile device users.

Area of Science:

• Computational health
• Behavioral science
• Machine learning

Background:

• Mobile devices enable self-monitoring of behaviors like exercise and health status like stress.
• Current devices lack support for understanding behavior-health status links.
• Personalized health interventions are needed to leverage self-monitoring data.

Purpose of the Study:

• Develop an interpretable policy for physical activity recommendations.
• Reduce user-perceived psychological stress over a defined time horizon.
• Introduce a novel sequential decision-making method.

Main Methods:

• Formulated the problem as a sequential decision-making task.
• Developed and applied threshold Q-learning (TQL), a new interpretable method.
• Analyzed data from 79 healthy adults over 7 weeks (physical activity and self-assessed stress).

Main Results:

• TQL demonstrated superior performance compared to traditional Q-learning under model misspecification.
• The TQL method is doubly robust and offers enhanced interpretability.
• Simulation results validated the effectiveness of TQL.

Conclusions:

• TQL provides an interpretable approach for personalized physical activity recommendations.
• This method can help users understand how their behavior impacts psychological stress.
• The study is a foundational step towards mobile-based computational health coaching.