Related Experiment Video
Updated: Jun 20, 2026
06:49
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
Published on: December 11, 2015
Evaluating Representation Embeddings from LLMs and Time-Series Foundation Models for Wearable Accelerometer-Based
1Department of Biostatistics, Columbia University Mailman School of Public Health, New York, NY.
Summary
Simple wearable sensor features effectively predict health outcomes, matching or exceeding complex AI models. Advanced AI embeddings offer minimal gains, highlighting the value of basic data variability for personalized health monitoring.
Area of Science:
- Biomedical Engineering
- Digital Health
- Wearable Technology
Background:
- Wearable accelerometers collect detailed behavioral data for personalized health insights.
- Limited comparative studies exist on modern representation-learning methods for accelerometer data.
Purpose of the Study:
- To compare the effectiveness of different data representation methods for predicting clinical outcomes using accelerometer data.
- To evaluate simple entropy-based features, large-language-model (LLM) embeddings, and time-series foundation model embeddings.
Main Methods:
- Utilized accelerometer data from the National Health and Nutrition Examination Survey (NHANES).
- Assessed prediction accuracy for outcomes including overweight status, lipid biomarkers, glucose, arthritis, and cancers.
- Compared entropy-based features, LLM embeddings, and time-series foundation model embeddings.
Main Results:
- Entropy-based features performed comparably or better than embedding approaches across all outcomes.
- LLM embeddings provided only marginal predictive improvements (AUC difference of 0.01-0.05).
- Time-series foundation model embeddings showed minimal utility, and prompt-based LLM reasoning performed poorly.
Conclusions:
- Simple variability features derived from accelerometer data are highly effective for health outcome prediction.
- Current advanced AI embeddings offer limited advantages over basic features for this data.
- Future health sensing foundation models require domain-aligned pretraining for improved performance.
