Behavioral Modeling in Weight Loss Interventions.

Area of Science:

• Computational Social Science
• Machine Learning
• Behavioral Economics

Background:

• Designing human-agent systems requires models of agent responses to system changes.
• Obesity treatments struggle with long-term patient adherence despite lifestyle interventions.
• Personalized treatment approaches show promise for improving adherence.

Purpose of the Study:

• To develop a framework for predictive modeling to personalize treatments.
• To incorporate time-varying system states and motivational states into predictive models.
• To utilize qualitative social science models of behavior change.

Main Methods:

• Developed a predictive modeling framework using utility functions.
• Formulated the computation of the predictive model as a mixed-integer linear program (MILP).
• Introduced a novel formulation for Bayesian inference using empirical histograms as prior distributions.

Main Results:

• Validated the predictive framework using a weight loss intervention dataset.
• Demonstrated the model's predictive ability compared to standard machine learning approaches.
• The proposed model enables optimization, unlike standard machine learning methods.

Conclusions:

• The developed framework offers a novel approach to predictive modeling for personalized interventions.
• The MILP reformulation enables Bayesian inference with empirical histogram priors.
• The predictive model can be utilized for treatment optimization, addressing limitations of current methods.