Perbandingan Metode Ensemble Learning pada Klasifikasi Penyakit Diabetes

Abstract

Diabetes is a disease in the medical world characterized by high blood sugar levels in the sufferer. According to data from the World Health Organization (WHO), between 1980 and 2014, there was an increase in diabetes cases from 108 million to 422 million. Ensemble Learning, which is one of the Machine Learning paradigms, can be used to classify diabetes. In this study, 3 Ensemble Learning methods were compared, namely Bagging, Boosting, and Stacking on 3 datasets. The 3 datasets used were Pima Indians Diabetes, Frankfurt Hospital Diabetes, and Sylhet Hospital Diabetes. From the results of ensemble learning experiments conducted on the three datasets, it was found that the Boosting method could outperform the Bagging and Stacking methods. In dataset 1, the highest accuracy is 81.82% with Gradient Boosting, Extreme Gradient Boosting, and Cat Boosting models. In dataset 2, the highest accuracy is 99.25% using the Light Gradient Boosting model. While the highest accuracy in the third dataset is 100% using the Light Gradient Boosting and Cat Boosting models.