Feature Selection for Physical Activity Prediction Using Ecological Momentary Assessments to Personalize Intervention...

Area of Science:

• Digital Health
• Behavioral Science
• Machine Learning

Background:

• Health apps increasingly use behavior change techniques.
• Just-in-time adaptive interventions (JITAIs) leverage passive sensing for personalized support.
• Ecological momentary assessment (EMA) can capture user context but may increase burden.

Purpose of the Study:

• To optimize EMA feature sets for predicting physical activity (PA) enactment.
• To balance prediction accuracy with the number of EMA questions using machine learning.

Main Methods:

• 43 healthy adults completed daily EMA surveys for 3 weeks.
• EMA assessed motivational, volitional, stress, and mood variables.
• PA enactment was the outcome, predicted by EMA data using machine learning.

Main Results:

• Machine learning models achieved a mean AUC of 0.87 for PA prediction.
• Key predictors included self-efficacy, stress, planning, and perceived barriers.
• A concise set of EMA features accurately predicted PA engagement.

Conclusions:

• Limited EMA features effectively predict physical activity.
• This prediction accuracy enables tailored JITAIs for improved health outcomes.
• Optimized EMA can enhance the effectiveness of digital health interventions.