Authors:
Mustaqim Sidebang
1
;
Erna Budhiarti Nababan
2
and
Sawaluddin
3
Affiliations:
1
Master of Informatics Program, Universitas Sumatera Utara, Medan, Indonesia
;
2
Faculty of Computer Science and Information Technology, Universitas Sumatera Utara, Medan, Indonesia
;
3
Faculty of Mathematics and Natural Sciences, Universitas Sumatera Utara, Medan, Indonesia
Keyword(s):
K-Means Algorithm, Pillar Technique, Determining Centroids.
Abstract:
The K-Means algorithm is a popular clustering technique used in many applications, including machine learn ing, data mining, and image processing. Despite its popularity, the algorithm has several limitations, including sensitivity to the initialization of centroid values and the quality of clustering. In this paper, we propose a novel technique called the ”pillar technique” to improve the performance of the K-Means algorithm. The pillar technique involves dividing the dataset into smaller sub-datasets, computing the centroids for each sub dataset, and then merging the centroids to obtain the final cluster centroids. We compare the performance of the K-Means algorithm with and without the pillar technique on several benchmark datasets. Our results show that the pillar technique improves the quality of clustering and convergence rate of the algorithm while reducing computational complexity. We also compare our proposed approach with other centroid initializa tion methods, including K-
Means++, and demonstrate the superior performance of the pillar technique. Our findings suggest that the pillar technique is an effective method to improve the performance of the K-Means algorithm, especially in large-scale data clustering applications.
(More)