An Accurate Glucose Detection Platform Using Colorimetry and Supervised Learning Algorithms.

Maintaining optimal health and preventing diabetes-related complications require accurate and timely monitoring of blood glucose levels. In this context, the present study develops an affordable, reliable, and precise Point-of-Care (POC) diagnostic platform for glucose detection by integrating microfluidic and colorimetric principles. The system uses a custom-fabricated microfluidic chip that enables efficient enzymatic color reactions with only ~20 µL of sample per microwell, achieving complete color development within 3-4 minutes. The chip is enclosed in a compact, USB-powered, 3D-printed imaging module equipped with a high-resolution fixed-focus camera to ensure stable control of focal distance, alignment, and illumination. The workflow is designed for seamless compatibility with embedded systems or laptops, eliminating dependence on smartphones or external calibration tools and supporting real-time POC deployment. A dataset of 1280 images representing 16 glucose concentrations from 50 to 200 mg/dL was captured under standardized conditions, labelled, and uniformly preprocessed. Engineered image features extracted from the processed images were evaluated using supervised machine learning models, including Random Forest, Support Vector Machine, K-Nearest Neighbours, and a Feedforward Neural Network, to create a robust predictive framework for rapid and consistent glucose estimation. Among these, the Random Forest model achieved the highest cross-validation precision of 98% and specificity approaching 100%, effectively distinguishing glucose levels. Confusion matrix and ROC analyses further confirmed reliability, showing minimal misclassification and a mean AUC near 1. Overall, the proposed image-based glucose estimation approach demonstrates a cost-effective, scalable, and accurate solution for real-time monitoring in diverse healthcare settings. The system also offers low reagent consumption, rapid analysis time, and straightforward operation, making it suitable for decentralized screening, routine monitoring, and resource-limited clinical environments where conventional laboratory testing may be inaccessible or delayed. Future work will focus on expanding concentration ranges, validating performance with clinical samples, and integrating automated calibration to support large-scale and long-term usability.