Predicting Adherence to Behavior Change Support Systems Using Machine Learning: Systematic Review.

Area of Science:

• Digital Health
• Machine Learning Applications
• Behavioral Science

Background:

• Limited reliable adherence prediction measures exist for behavior change support systems (BCSSs).
• Existing reviews focus on self-reporting, which is prone to inaccuracies in adherence behavior.
• There is a need for objective and accurate methods to predict adherence.

Purpose of the Study:

• To systematically review and summarize machine learning (ML) approaches for predicting adherence to BCSSs.
• To identify trends in ML applications for adherence prediction.
• To assess the effectiveness of ML models in this domain.

Main Methods:

• Systematic literature search of Scopus and PubMed (January 2011 - August 2022).
• Inclusion of 11 eligible studies from an initial retrieval of 2182 papers.
• Analysis of identified machine learning techniques and adherence categories.

Main Results:

• Four adherence categories identified: digital interventions, medication, physical activity, and diet.
• Machine learning for real-time adherence prediction is a growing research area.
• 13 supervised learning techniques used, mostly traditional (e.g., support vector machine); advanced techniques include LSTM, multilayer perception, and ensemble learning.
• Most models achieved good classification accuracy, indicating effective feature selection.

Conclusions:

• Machine learning algorithms can predict user adherence in BCSSs.
• Predictive models facilitate adherence behavior reinforcement.
• Development of intelligent BCSSs with personalized, timely suggestions is enabled by ML.