Developing and comparing a new BMI inclusive energy expenditure algorithm on wrist-worn wearables

Summary: We present a BMI-inclusive algorithm for estimating energy expenditure from wrist sensors and validate its robustness across diverse BMI groups.

Abstract

Accurate wrist-worn energy expenditure (EE) estimation often degrades for higher or lower BMI populations due to distribution shift in motion and cardiovascular signals. We introduce a BMI-inclusive EE framework that calibrates feature distributions across BMI strata, combines activity-specific models with global regularization, and preserves real-time efficiency for wearables. Evaluations on multi-BMI cohorts show improved agreement against reference EE with reduced bias across BMI groups while maintaining low latency suitable for on-device inference.