Dian Septiana, Sisti Nadia Amalia, Fanny Ramadhani, Fahmi Ashari S Sihaloho, Prihatin Ningsih Sagala
This study proposes a data-driven approach to classify the economic potential of 33 districts and cities in North Sumatra, Indonesia, using unsupervised machine learning. A total of 23 indicators covering economic, demographic, and spatial dimensions were standardized and reduced using Principal Component Analysis (PCA), retaining over 95% of the total variance. Clustering was performed using K-means, DBSCAN, and agglomerative clustering. Internal validation with the Silhouette Score and Davies-Bouldin Index (DBI) showed that agglomerative clustering with three clusters achieved the best performance, with a Silhouette Score of 0.313 and DBI of 0.871. External evaluation using the Adjusted Rand Index (ARI) indicated that KMeans achieved the highest structural alignment. Spatial mapping revealed that the three-cluster agglomerative model outlined macrozones such as urban cores, transitional interiors, and peripheral islands, whereas the five-cluster KMeans model provided finer distinctions by identifying urban-industrial corridors, highland tourism areas, and emerging servicebased municipalities. DBSCAN exposed structural ambiguity, with a four-cluster configuration mirroring KMeans patterns but highlighting uncertain regions through a 36.4 % noise rate, whereas a three-cluster setting captured broader groupings with lower noise. These results demonstrate that unsupervised learning can produce flexible and interpretable regional classifications to support adaptive and evidence-based spatial planning. © 2025 IEEE.
Universitas Negeri Medan, Department of Statistics, Medan, Indonesia; Universitas Negeri Medan, Department of Computer Science, Medan, Indonesia; Universitas Negeri Medan, Department of Economics Education, Medan, Indonesia; Universitas Negeri Medan, Department of Mathematics Education, Medan, Indonesia
