Peningkatan Keberagaman Data untuk Klasifikasi Penyakit Diabetes Berbasis Stacking Ensemble Learning

Abstrak

Diabetes cases are becoming more common in the late years. Diabetes attacks not only parents, but also children. The development of diabetes is not far from the lifestyle and diet that we live on a daily basis. Therefore, early detection of diabetes is essential because the earlier the disease is detected, the easier it is to treat. In the process of detecting disease based on factors, the cause can be predicted with data mining. The aim of this research is to increase data diversity so that it can be processed to the maximum in data mining. In the process of data upgrading, we used the imbalance learning method SMOTE-ENN combined with the method Stacking Ensemble Learning. In the search for a powerful stacking model, seven classification algorithms were involved in the experiments carried out on this study, namely: Random Forest, Decision Tree, Gradient Boosting, Naïve Bayes, Extreme Gradiant Boost, Logistic Regression, and k-Nearest Neighbor. Four algorithms were used to be classifiers level 0 (base model), namely kNN, Gradient Boosting, decision tree, and random forest, while Random Forest was used again to be classifier level 1. (meta model). With these combinations, the accuracy obtained is 97.3%. These are the highest results when compared to individual algorithms.