GlucoFM-Bench: Benchmarking Time-Series Foundation Models for Blood Glucose Forecasting

2026-06-04

AI Summary

This study addresses the lack of systematic evaluation of temporal foundation models (TSFMs) for blood glucose prediction across diverse populations and varying data regimes. We introduce GlucoFM-Bench, the first standardized benchmark for glucose forecasting, encompassing 15 diabetes datasets and evaluating eight TSFMs—including Chronos-2 and TimesFM—alongside specialized models like LSTM under zero-shot, few-shot, and full-data training scenarios. Our results demonstrate that TSFMs achieve near-supervised performance in zero-shot settings (within 5% of fully supervised baselines), yet lightweight LSTMs significantly outperform TSFMs by 4–21% when sufficient data is available. The evaluation further reveals persistent challenges in predicting glucose levels for type 1 diabetes patients and extreme glycemic ranges, underscoring the necessity of granular, context-aware assessment protocols.

Abstract

Blood glucose forecasting models are foundational for modern diabetes management systems, as reliable short-term predictions can enable proactive interventions, support automated insulin delivery, and reduce the risk of hypo- and hyperglycemic events. From a modeling perspective, glucose forecasting poses unique challenges due to heterogeneous physiological dynamics across diabetes populations. Traditional machine learning and deep learning models have been extensively evaluated for glucose prediction, yet recent time-series foundation models (TSFMs) remain much less studied in this setting. To bridge this gap, we present GlucoFM-Bench, a comprehensive benchmark evaluating state-of-the-art TSFMs alongside supervised deep learning models for blood glucose forecasting. We assess eight representative architectures, including pre-trained TSFMs, time-series large language models, and task-specific deep learning models, across 15 publicly available diabetes-relevant datasets comprising 1,117 individuals with type 1 diabetes, type 2 diabetes, prediabetes, and no diabetes. Models are evaluated under zero-shot, few-shot, and full-shot protocols, with systematic variation in context length and prediction horizon. Across datasets, pre-trained TSFMs, especially Chronos-2 and TimesFM, show strong zero-shot and few-shot transfer, with the best zero-shot model performing within 5% of the best full-shot supervised model. Yet, when task-specific data are abundant, a lightweight LSTM remains strongest, outperforming TSFMs by 4--21% under full-shot training. Stratified analyses reveal persistent challenges in T1D cohorts and hypo-/hyperglycemic ranges, highlighting the need for evaluation beyond aggregate error metrics. Together, GlucoFM-Bench provides a standardized and reproducible foundation for evaluating, comparing, and improving foundation models for blood glucose forecasting.

Problem

• blood glucose forecasting
• time-series foundation models
• diabetes management
• model benchmarking
• predictive modeling

Innovation

• time-series foundation models
• blood glucose forecasting
• zero-shot learning
• benchmarking
• diabetes management