Weighted Ensemble Based Noninvasive HbA1c Estimation from PPG signals Using Machine Learning Algorithms

Noninvasive assessment of glycated hemoglobin (HbA1c) levels is an ongoing area of research in diabetes management, providing a patient-friendly alternative to traditional blood-based measurements. In this study, we introduce an approach to estimate HbA1c from photoplethysmography (PPG) signals by applying a weighted ensemble of machine learning models, increasing accuracy and improving prediction reliability by prioritizing models with lower mean squared error (MSE). Machine learning algorithms such as random forest and various boosting algorithms such as XGBoost, LightGBM, CatBoost, gradient boosting and AdaBoost were used to estimate HbA1c levels from wrist PPG signals.