Paper
Data leakage in PPG-based non-invasive blood glucose prediction
RCMLR 2026 poster (OpenReview AQu6vzPJhc, 2026-09-30)
TL;DR
Re-evaluates three PPG blood-glucose models (TinyML, CBHFF, CatBoost) on three datasets under five protocols: random split, subject-wise random split, participant-aware split, leave-some-participants-out and cross-dataset OOD. Random splits give the best numbers; performance falls sharply under subject separation, and cross-dataset OOD gives near-zero or negative R², while Clarke Error Grid A+B stays at 97–100% even when regression fails.
- Form factor: mixed
Results on ChatPPG leaderboards
All leaderboardsNo results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.
Tasks & signals
Tasks
Signals
Keywords
How it works
Cite this paper
Shen, J., Liu, H., Liesaputra, V., Zhang, Y., & Huang, Z. (2026). Data leakage in PPG-based non-invasive blood glucose prediction. RCMLR 2026 poster. https://openreview.net/forum?id=AQu6vzPJhc
Shen, Junyi, Hongmei Liu, Veronica Liesaputra, Yuan Zhang, and Zhiyi Huang. 2026. "Data leakage in PPG-based non-invasive blood glucose prediction." RCMLR 2026 poster. https://openreview.net/forum?id=AQu6vzPJhc.
Shen, J., Liu, H., Liesaputra, V., Zhang, Y. and Huang, Z. (2026) 'Data leakage in PPG-based non-invasive blood glucose prediction', RCMLR 2026 poster. Available at: https://openreview.net/forum?id=AQu6vzPJhc
Shen, Junyi, et al. "Data leakage in PPG-based non-invasive blood glucose prediction." RCMLR 2026 poster, 2026. https://openreview.net/forum?id=AQu6vzPJhc.
J. Shen, H. Liu, V. Liesaputra, Y. Zhang, and Z. Huang, "Data leakage in PPG-based non-invasive blood glucose prediction," RCMLR 2026 poster, 2026, [Online]. Available: https://openreview.net/forum?id=AQu6vzPJhc.
@article{shen2026data,
title = {{Data leakage in PPG-based non-invasive blood glucose prediction}},
author = {Shen, Junyi and Liu, Hongmei and Liesaputra, Veronica and Zhang, Yuan and Huang, Zhiyi},
year = {2026},
journal = {RCMLR 2026 poster},
url = {https://openreview.net/forum?id=AQu6vzPJhc}
}Added to the ChatPPG research hub