Peningkatan Akurasi K-Means menggunakan Kombinasi Rapid Centroid Estimation dan Canberra Distance

Abstract

K-Means is a widely used clustering method due to its simplicity; however, it has limitations related to the random initialization of centroids and its reliance on the Euclidean Distance metric. This study aims to improve the accuracy of K-Means by integrating the Rapid Centroid Estimation (RCE) method for initial centroid selection and employing the Canberra Distance as the distance metric. The number of clusters is determined using the Elbow method, and clustering performance is evaluated using the Silhouette Coefficient. Experiments were conducted on two datasets: Wholesale Customers and New Student Enrollment Data. The results show that the combination of RCE and Canberra Distance in K-Means yields significantly improved clustering accuracy. At the optimal number of clusters (k = 3), accuracy increased from 33.30% to 58.40% on the Wholesale Customers dataset, and from 25.51% to 34.61% on the New Student dataset after applying Min-max Normalization. The proposed approach demonstrates its effectiveness in producing higher-quality clustering compared to the standard K-Means without the integration of RCE and Canberra Distance.