Hybrid grey wolf optimization and machine learning models for stunting malnutrition classification

Abstract

This study investigates the impact of Synthetic Minority Oversampling Technique (SMOTE) and feature selection on ten machine learning models hybridized with Grey Wolf Optimization (GWO) for stunting malnutrition classification. The models, including Support Vector Classifier (GWO+SVC), KNearest Neighbors (GWO+KNN), Decision Tree (GWO+DT), Gradient Boosting Decision Trees (GWO+GBDT), Random Forest (GWO+RF), XGBoost (GWO+XGBoost), LightGBM (GA+LightGBM), Adaboost (GWO+Adaboost), CatBoost (GWO+CatBoost), and Stacking Classifier (GWO+Stacking), are evaluated. The findings reveal that SMOTE significantly enhances most models, particularly in terms of accuracy, precision, and F1-score, effectively addressing data imbalance. Notably, GWO+GBDT consistently outperforms other models, achieving the highest accuracy after SMOTE application. Feature selection further boosts model performance, with GBDT reaching 86.84% accuracy after attribute selection. Future research avenues may explore alternative machine learning models, feature selection methods, additional data sources, external validation, parameter optimization, and the integration of clinical information for more effective stunting malnutrition combat strategies.