Personalising Digital Health Behaviour Change Interventions using Machine Learning and Domain Knowledge

TL;DR

This paper presents a virtual coaching system that personalizes digital health interventions by predicting patient behavior and using counterfactuals for tailored guidance, tested with simulated data.

Contribution

It introduces a novel system combining machine learning and domain knowledge to personalize health behavior interventions using counterfactual explanations.

Findings

System can predict patient adherence to interventions

Counterfactual examples improve personalization

Simulated data supports system evaluation

Abstract

We are developing a virtual coaching system that helps patients adhere to behavior change interventions (BCI). Our proposed system predicts whether a patient will perform the targeted behaviour and uses counterfactual examples with feature control to guide personalisation of BCI. We use simulated patient data with varying levels of receptivity to intervention to arrive at the study design which would enable evaluation of our system.