Analisis Metode Elbow SSE, Silhouette Score, dan Jaccard Stability dalam Pemilihan Jumlah Klaster Data yang Optimal

Hartono, Budi
Lusiana, Veronica

TIN: TERAPAN INFORMATIKA NUSANTARA
Vol 6 No 8 (2026): January 2026

This study discusses the selection of the optimal number of clusters (K) in the K-Means algorithm by utilizing a combination of the Elbow method with the SSE (Sum of Squared Errors) and Silhouette Score metrics. The main problem is that the optimal K value is unknown. Choosing K that is too small can combine different patterns (under-clustering), and choosing K that is too large can break the same pattern into several clusters (over-clustering). The experiment used two-dimensional test data with variations in the number of data 20, 30, 40, 50, and 60. K-Means was run in the range of K = 2 to K = 8, then the SSE value was calculated to form the Elbow curve and the average Silhouette value to evaluate the quality of the cluster. This study added a cluster stability test using the Jaccard Stability value. The highest Silhouette value of 0.4619 was obtained from the data 20 for K = 2. The highest Jaccard stability value of 0.9507 was obtained from 60 data sets for K = 2. The experimental results show that the Elbow method, Silhouette value, and Jaccard stability can be used complementarily in determining the optimal K. In some test data, both metrics produce consistent K recommendations, while in certain test data, Elbow can provide several candidates, so that validation using the Silhouette value is needed to select the optimal K.