Abstract
Stunting remains a critical nutritional issue among children, affecting growth and long-term human resource quality. Despite national programs and global targets for stunting reduction, early prediction of stunted children using data-driven methods remains limited. This study aims to evaluate and compare the performance of four supervised machine learning algorithms—Naïve Bayes, Multilayer Perceptron (MLP), Decision Tree (J48), and Support Vector Machine (SVM)—in predicting stunting using a dataset of 97 child records from three villages in East Kalimantan, Indonesia. Data were tested in both unnormalized and normalized forms and split into training and testing sets at 70%–30%, 80%–20%, and 90%–10% ratios. The results indicate that MLP and Decision Tree consistently achieved 100% accuracy across all splits and preprocessing conditions, while Naïve Bayes and SVM showed lower and more variable accuracy in certain cases. These findings suggest that MLP and Decision Tree are the most reliable methods for stunting prediction in small datasets, providing a practical approach for early identification and intervention. The study highlights the importance of algorithm selection and preprocessing in achieving optimal predictive performance in health-related datasets.