<![CDATA[Stunting is one of the most prevalent chronic nutritional disorders affecting toddlers, with significant impacts on their physical growth and cognitive development. The limited quality and completeness of health records collected manually at community health posts (Posyandu) present challenges for the early identification of stunting. This study aimed to develop a predictive model for stunting levels in toddlers using a machine learning approach based on health data and image processing. The theoretical framework incorporated Machine Learning, Support Vector Machine (SVM), Random Forest, Convolutional Neural Network (CNN), Image Processing, and Ensemble Learning using the Hard Voting technique. This study employed a quantitative experimental design following the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. The data were obtained through observations and the collection of child health records at the Posyandu in Lubuk Sahung Village, Seluma Regency, Bengkulu Province, complemented by a public dataset from Kaggle and body images of children under five. The research process included data preprocessing, image feature extraction using CNN, feature fusion of health data and image features, model training using SVM and Random Forest, and model integration through Ensemble Voting. The results demonstrated that the SVM and Ensemble Voting models achieved the best predictive performance, with an accuracy of 91.67%, precision of 100.00%, recall of 84.62%, and an F1-score of 91.67%, while the Random Forest model achieved an accuracy of 83.33%. Based on the generalization analysis, the SVM model exhibited the highest stability, with a training–testing accuracy gap of only 4.16%. These findings indicate that integrating health data with image-derived features can effectively support the prediction of stunting levels and serve as a reliable decision-support tool for the early monitoring of children's nutritional status.]]>
Copyrights © 2026
