Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{209082,
        author = {Vedika Shinde and Arya Balpande and Tanisha Nipunge},
        title = {Machine Learning Based Blood Glucose Estimation from PPG Signals},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {5},
        pages = {741-756},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=209082},
        abstract = {Diabetes mellitus is a chronic metabolic disorder affecting over 537 million adults worldwide, with projections exceeding 780 million by 2045. Conventional blood glucose monitoring depends on invasive finger-prick capillary sampling, which causes discomfort and reduces patient adherence. This paper presents a low-cost, non-invasive blood glucose level (BGL) estimation system that acquires photoplethysmography (PPG) signals using an analog infrared (IR) sensor connected to an Arduino microcontroller, and applies supervised machine learning for real-time regression-based BGL prediction.
Raw 10-bit ADC PPG signals are sampled at 100 Hz over 180 seconds, band-pass filtered using a 4th-order Butterworth filter (0.5–10 Hz), and a 21-dimensional feature vector spanning time-domain, frequency-domain, morphological, and patient-metadata categories is extracted. A Gradient Boosting Regressor (GBR) and a Random Forest Regressor (RFR) are evaluated via 5-fold cross-validation on 67 recordings from 23 subjects (BGL 88–183 mg/dL, mean 115.0 ± 18.6 mg/dL). The GBR achieved lower MAE (±8–15 mg/dL, R² = 0.72–0.85) and was deployed as the production model. An enhanced four-criterion signal quality gate — including a cardiac-band spectral check — prevents spurious predictions from corrupted acquisitions.

Live validation on 22 subjects yielded BGL predictions of 122.7–132.8 mg/dL, all correctly classified as Normal, with two acquisitions correctly rejected. The complete system is delivered as a real-time desktop application built with CustomTkinter, featuring live PPG waveform display, automated signal quality gating, per-patient trend tracking, returning-patient detection, and a persistent records manager. Total hardware cost is under USD 6.},
        keywords = {photoplethysmography (PPG), non-invasive blood glucose monitoring, gradient boosting regression, random forest, signal processing, machine learning, diabetes, Arduino, CustomTkinter.},
        month = {October},
        }