<![CDATA[Background: Community-based child-growth monitoring requires reliable record management and transparent classification. Objective: This study documents a web-based Posyandu prototype for Mekar Jaya Village and replicates a four-class anthropometric classification experiment using an identifiable public dataset. Methods: The prototype used Laravel 10 and MySQL. A Kaggle dataset (version 2; 120,999 records) was analyzed using age, sex, and height. Stratified group-based splitting yielded 96,909 training and 24,090 untouched test records; identical predictor vectors were grouped, and random oversampling was restricted to training data. Random Forest (200 trees) and Gaussian Naive Bayes were compared on the same test set. Results: Random Forest achieved 98.89% accuracy (95% bootstrap CI: 98.76–99.02%), macro-F1 98.49%, and macro-recall 98.42%; Gaussian Naive Bayes achieved 49.63%, 41.42%, and 43.36%, respectively. Among 2,755 stunted test records, 40 were classified as Normal; none of the 3,961 severely stunted records was classified as Normal. Conclusion: The replication provides auditable evidence for reproducing dataset labels, while the web prototype demonstrates an application interface. The new results are not a verification of previously reported accuracies or evidence of local clinical validity; label provenance, local validation, and software acceptance testing remain necessary.]]>
Copyrights © 2026
