Analisis Performa SMOTE, ADASYN, dan Random Oversampling pada Klasifikasi Penyakit Stunting dengan Algoritma CatBoost

Stunting is a chronic nutritional problem influenced by inadequate nutritional intake, suboptimal parenting, and social and environmental factors. This study aims to classify stunting using the CatBoost algorithm with three data balancing methods: SMOTE, ADASYN, and Random Oversampling. The results showed that ADASYN provided the best performance with an accuracy of 85%, followed by SMOTE with 84%, while Random Oversampling produced the lowest accuracy of 73%. Additional experiments using explainability techniques (Feature Importance, SHAP Values, and Permutation Importance) identified five key features: Age, Birth Weight, Birth Length, Body Weight, and Body Length. Re-evaluation with only these five features showed no significant change in performance compared to the model using all features. These findings demonstrate that the combination of appropriate balancing techniques and explainability analysis can produce a model that is not only accurate but also efficient in detecting potential stunting.