Integration of K-Means Clustering and K-Nearest Neighbor for Mapping and Predicting Stunting Risk Among Toddlers in Indonesia

Abstract

 Stunting prevalence among children under five in Indonesia keeps falling, yet remains unevenly spread, so setting priority areas needs a verifiable quantitative basis. Previous studies generally use a single algorithm, so the resulting map yields no estimate for the next period. This study combines both in one workflow: K-Means Clustering groups 514 districts and cities by risk level, and the cluster labels become reference classes for K-Nearest Neighbor to predict the 2025 risk category. The data are under-five stunting prevalence from the Indonesian Nutritional Status Survey (SSGI) 2021–2024. Its 128 empty values were filled by mean computation and rescaled with Min-Max normalisation, distance used Euclidean Distance, cluster quality the Silhouette Coefficient, and classification 23 nearest neighbours from a 1,542-row training set. Iteration stopped at the fifteenth round, giving 60 districts in High Risk, 232 in Moderate, and 222 in Low Risk, with a mean Silhouette Score of 0.33. The 2025 prediction places 25 districts in High Risk, 194 in Moderate, and 295 in Low Risk, agreeing with the cluster labels on 378 of 514 districts, or 73.54%. Re-running it in Orange Data Mining differed by at most four districts, confirming the workflow and its map for setting intervention priorities.