Analysis of Stunting Prediction in Toddlers in Bekasi District Using Random Forest and Naïve Bayes

Abstract

This study aims to compare the performance of the Random Forest and Naïve Bayes algorithms in predicting stunting in toddlers using data from the Bekasi District Health Office. The analysis process begins with data cleaning, normalization, and sampling using the Adaptive Synthetic Sampling (ADASYN) method to handle data imbalance, followed by validation with Stratified K-Fold Cross Validation. The implementation of the algorithm shows that Random Forest has the highest accuracy of 89.62% and an F1-Score of 89.09%. Naïve Bayes Gaussian produces an accuracy of 88.72% and an F1-Score of 88.81%, while Naïve Bayes Bernoulli has a lower performance with an accuracy of 67.83% and an F1-Score of 69.72%. Random Forest shows advantages in overcoming noise and imbalanced data, making it an optimal choice for stunting prediction. Meanwhile, the performance of Naïve Bayes is influenced by the characteristics of the data, where the Gaussian variation is more suitable for continuous data. The results of this study provide insight that choosing the right algorithm, especially on imbalanced data, is very important to improve prediction accuracy. This study also recommends more attention to data preprocessing to ensure optimal prediction quality, especially for minority classes.