Machine Learning-Based Modeling of Blood Glucose Levels from Photoplethysmography Signals

This report presents a comprehensive study on non-invasive blood glucose estimation using Photoplethysmography (PPG) signals combined with machine learning techniques. The research is conducted within the Structured Engineering Thesis Framework (SETF), a domain-aware and workflow-driven methodology that ensures a systematic and reproducible research process. The work integrates principles from biomedical signal processing, artificial intelligence, and software engineering to develop an end-to-end system for glucose level prediction. It covers all stages of the research lifecycle, including problem definition, literature review, dataset preparation, signal preprocessing, feature extraction, model selection, validation, and evaluation. Special attention is given to critical methodological aspects such as prevention of data leakage, selection of appropriate evaluation metrics, and ensuring model generalizability. The study also incorporates system design considerations and practical implementation aspects, including smartphone-based data acquisition using PPG signals. By combining theoretical foundations with applied machine learning models, this research contributes to the advancement of non-invasive glucose monitoring technologies. The results demonstrate the feasibility of using PPG signals as a reliable indicator for estimating blood glucose levels, supporting the development of accessible and cost-effective healthcare solutions. This work was conducted at Arab International University (AIU), Syria. The official website of the university is: https://www.aiu.edu.sy