K-Means Clustering Optimization Using the Elbow Method and Early Centroid Determination Based on Mean and Median Formula
Edy Umargono, Jatmiko Endro Suseno, S K Vincensius Gunawan
Abstract
The most widely used algorithm in the cluster partitioning method is the K-Means algorithm, K-Means is an iteration algorithm with the user determining the number of clusters that need to be grouped and determining the centroid for each cluster so that the level of similarity between members in one group is high while the level of similarity with members in other groups is very low. Historically K-Means is still the best grouping algorithm among other grouping algorithms with the ability to group a number of data with relatively fast and efficient computing time. The K-Means algorithm is widely implemented in various fields in industrial and scientific applications and is very suitable for processing quantitative data with numeric attributes, but there are still weaknesses in this algorithm. Weaknesses of the K-Means algorithm include determining the number of clusters based on assumptions and relying heavily on the initial selection of centroids to overcome this weakness, in this study, we propose the use of the elbow method to determine the best number of clusters and initials. Centroid determination based on average and median data. The results of this study indicate that using initial cluster center determination based on average data makes the number of iterations needed to achieve uniformity in clusters 22.58% less than initial random cluster determination and determining the best number of clusters using the elbow method makes the required iteration 25% less than using the number of other clusters.