PROJECTS

This project introduces a framework that uses wearable sensor data to generate personalized counterfactuals — answering “what if” questions about lifestyle changes and their potential health outcomes. By combining multimodal similarity analysis with a temporal PC algorithm, the system uncovers predictive relationships across time and trains gradient boosting models to estimate individual-specific effects. A counterfactual engine then projects physiological trajectories under hypothetical interventions, such as altered activity or sleep patterns. Evaluations show strong predictive accuracy and high counterfactual plausibility, highlighting significant inter-individual variability. This framework offers a powerful tool for exploring personalized health dynamics and tailoring interventions to individual needs.