Hybrid DAC-GA and K-Means for Spatial Clustering of Stunting Risk in North Sumatra

Abstract

Stunting continues to pose a severe global health concern, particularly in Indonesia, where prevalence rates persist above international standards despite recent advances in reduction initiatives. Accurately documenting the regional variation of stunting is critical to facilitate targeted interventions and successful policymaking. This paper offers a hybrid clustering framework that merges the classic K-Means approach with the Dynamic Artificial Chromosomes Genetic approach (DAC-GA) to increase the resilience and reliability of spatial analysis. The dataset used combines demographic and population statistics from the Central Bureau of Statistics (BPS), strategic policy documents from the Regional Medium-Term Development Plan (RPJMD) of North Sumatra, and health indicators including stunting prevalence data from the Ministry of Health of the Republic of Indonesia.

The research approach consists of four primary phases: data preparation, clustering model construction, cluster evaluation, and geographical visualization. Three evaluation metrics Sum of Squared Errors (SSE), Davies–Bouldin Index (DBI), and Silhouette Coefficient were applied to validate clustering performance. Results demonstrate that DAC-GA dynamically determined the ideal number of clusters at k=2 in just 1.171677 seconds, classifying Kota Medan and Deli Serdang into the low-risk cluster, while all other districts were consistently put into the high-risk cluster. Both DAC-GA and standard K-Means yielded similar spatial maps, giving significant methodological validation and strengthening the dependability of the findings. The study reveals not just the technical advantages of DAC-GA in maximizing clustering but also its practical utility in guiding spatially targeted health interventions. Future research is recommended to add dimensionality reduction utilizing Principal Component Analysis (PCA) to improve computing efficiency and enhance the interpretability of clustering results.