A MACHINE LEARNING APPROACH FOR PREDICTING STUNTING RISK IN TODDLERS

Abstract

Stunting is a chronic nutritional problem that remains a major public health challenge, particularly in developing countries such as Indonesia. It results from long-term nutritional deficiencies and can negatively affect physical growth, cognitive development, educational achievement, and future productivity. Early detection of stunting risk is essential to support timely intervention and improve child health outcomes. This study aims to develop and compare the performance of several machine learning algorithms for predicting stunting risk in toddlers using a large-scale nutritional dataset. The dataset was obtained from the Kaggle repository entitled “Stunting Balita Detection (121K Rows)” and consists of 120,999 records containing age, gender, height, and nutritional status information. Data preprocessing included categorical data encoding, Min-Max normalization, and dataset partitioning into training and testing sets using an 80:20 ratio. Five classification algorithms were evaluated: K-Nearest Neighbor (KNN), Random Forest, Support Vector Machine (SVM), Naïve Bayes, and Decision Tree C4.5. Model performance was measured using confusion matrix analysis, accuracy, precision, recall, and F1-score. The experimental results showed that KNN achieved the highest performance with an accuracy of 99.94%, precision of 99.90%, recall of 99.93%, and F1-score of 99.92%. Random Forest achieved comparable results with an accuracy of 99.93%, while SVM, Decision Tree C4.5, and Naïve Bayes produced lower performance values. These findings indicate that KNN and Random Forest are highly effective for stunting risk classification and have strong potential to support intelligent decision-support systems for early detection and nutritional monitoring of toddlers.