Synergistic Modeling of PPG Morphophysiological Patterns and Clinical Parameters for Fine-Grained Glycemic State Inference

Abstract

The increasing prevalence of diabetes and related metabolic disorders has created a strong demand for intelligent systems capable of accurately predicting blood glucose levels using non-invasive physiological signals. Continuous monitoring and timely prediction are essential to prevent complications and support effective clinical decision-making. However, many earlier approaches rely on basic statistical analysis and manual interpretation, which struggle to handle complex relationships among multiple health parameters such as Photoplethysmography (PPG) signal characteristics, heart rate, and patient demographics. These methods often lack scalability, adaptability, and predictive accuracy when applied to real-world healthcare datasets. Additionally, the absence of automated systems limits real-time analysis and delays preventive interventions. To address these challenges, this project proposes a machine learning-based predictive framework that leverages multiple regression models, including Linear Regression (LR), Ridge Regression, Least Absolute Shrinkage and Selection Operator (Lasso) Regression, Decision Tree Regressor (DTR), and an ensemble-based approach using Natural Gradient Boosting (NGB). The system processes physiological signals such as PPG along with attributes like age, gender, height, and weight to estimate glucose levels. Data preprocessing techniques such as missing value handling, feature scaling, and train-test splitting are applied to ensure robust model performance. The framework is deployed through a Flask based web application that enables model training, performance evaluation using metrics like Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R²) score, and real-time prediction for both single and batch inputs. The proposed system enhances prediction accuracy, reduces manual dependency, and provides a scalable solution for healthcare monitoring, enabling early detection and improved patient outcomes