Medilla Kusriyanto, Moh. Khairudin, Rustam Asnawi
Diabetes mellitus is one of the most prevalent chronic diseases worldwide, and accurate blood glucose level prediction plays a crucial role in early detection and effective management. Conventional machine learning approaches, such as Random Forest or Support Vector Machine (SVM), often face challenges related to scalability, efficiency, and generalization. In this study, we propose a dual predictive framework based on the Light Gradient Boosting Machine (LightGBM) algorithm to perform classification and regression tasks in blood glucose analysis. This dataset includes key biometric features such as Age, BMI, PPG data, HbA1C, fasting blood glucose level, and calories. LightGBM achieved an accuracy of 82.28 % with an F1 score of 0.85 for diabetes classification, 0.95 for normal classification, and 0.61 for prediabetes classification. This indicates that the LightGBM algorithm is able to classify blood sugar levels into 3 classes: normal, prediabetes, and diabetes, supported by confusion matrix analysis and feature importance evaluation. In predicting HbA1C levels, the LightGBM algorithm obtained an MAE of 0.9352 with an R2 of 0.4907. Although LightGBM's analysis of prediction results is inferior to Random Forest, it has better classification performance than Random Forest and SVM. © 2025 IEEE.
State University of Yogyakarta, Doctoral Study Program in Engineering Science Engineering Faculty, Indonesia; Islamic University of Indonesia, Industrial Faculty of Technology, Electrical Engineering Department, Indonesia
