Idiographic Dynamic Modeling for Behavioral Interventions with Mixed Data Partitioning and Discrete Simultaneous...

Area of Science:

• Behavioral Science
• Computational Modeling
• Personalized Interventions

Background:

• Personalized behavioral interventions require accurate dynamic models for individual subjects.
• Existing methods for model feature and parameter estimation can be computationally intensive.
• Optimizing interventions like physical activity promotion necessitates efficient modeling techniques.

Purpose of the Study:

• To present Discrete Simultaneous Perturbation Stochastic Approximation (DSPSA) as an efficient routine method for idiographic dynamic model development.
• To evaluate DSPSA's effectiveness in determining model features and parameters for personalized behavioral interventions.
• To compare DSPSA performance against exhaustive search methods using real-world intervention data.

Main Methods:

• Application of DSPSA for feature selection and regressor order determination in AutoRegressive with eXogenous input (ARX) models.
• Utilizing participant data from the 'Just Walk' physical activity intervention study.
• Employing various partitions of estimation and validation data to assess model robustness.

Main Results:

• DSPSA efficiently and rapidly estimated walking behavior models for individual participants.
• DSPSA proved effective in searching over model features and regressor orders, outperforming exhaustive search in speed.
• The study highlighted the importance of data partitioning strategies in idiographic modeling.

Conclusions:

• DSPSA is a valuable and efficient method for developing idiographic dynamic models for personalized behavioral interventions.
• Estimated models can inform the development of control systems to optimize intervention impact.
• Careful consideration of data partitioning is crucial for robust idiographic modeling.