Optimizing Stacked KNN, Naive Bayes, and LDA Models Using Random Forest as a Meta-Learner for Diabetes Classification

Meda, Ridodio Andreuw
Purwanto, Purwanto
Zami, Farrikh Al
Riyanto, Ahmad

Jurnal Penelitian Pendidikan IPA (JPPIPA)
Vol 11 No 10 (2025): October

Diabetes is one of the chronic diseases with a high mortality rate that requires proper treatment and early detection. This study proposes a stacking model approach with a combination of K-Nearest Neighbor (KNN), Naive Bayes, and Linear Discriminant Analysis (LDA) as the base-learner, and Random Forest as the meta-learner. The main objective of this study is to improve the classification accuracy of diabetes datasets that have an unbalanced class distribution. The experiment was conducted on the Pima Indians Diabetes dataset from the UCI Machine Learning Repository. The test results showed that the proposed stacking model was able to achieve an accuracy of 96.30%, True Positive Rate (TPR) of 88.89%, True Negative Rate (TNR) of 100%, and G-Mean of 94.28%. This performance is significantly better than the previous single classifier model and stacking approach. Thus, the proposed stacking model can be used as an effective solution in the classification of diabetic diseases under conditions of unbalanced class distribution.