Klasifikasi Multikelas Tingkat Diabetes Berdasarkan Indikator Kesehatan Pasien Menggunakan Strategi One-vs-Rest

Abstract

Diabetes is a non-communicable disease with a steadily increasing global prevalence. It often remains undiagnosed in its early stages, particularly during the prediabetic phase, which typically lacks noticeable symptoms. This study aims to develop a multi-class classification model to predict diabetes severity levels non-diabetic, prediabetic, and diabetic based on patient health indicators. A One-vs-Rest (OvR) strategy was employed, training each class against a combination of the others. The dataset was derived from the 2015 National Health Survey, comprising over 250,000 patient records with features such as blood pressure, body mass index, cholesterol levels, history of heart disease, and physical activity. Two machine learning algorithms, Logistic Regression and Random Forest, were applied to train the models. Class imbalance was addressed using the Synthetic Minority Over-sampling Technique (SMOTE). Evaluation metrics included accuracy, precision, recall, F1-score, and confusion matrix. The results show that the Random Forest model achieved an average accuracy of 93% and consistently high F1-scores, particularly for the prediabetic class of 98%. The most influential predictors were high blood pressure, obesity, and insufficient physical activity. This study contributes to the development of a reliable and efficient data-driven system for early diabetes risk detection.