Optimized K-Medoid: A Comprehensive Approach to Medoid Discovery and Finding Optimal K

Abstract

Clustering is an unsupervised learning algorithm used to partition data and has wide application in machine learning and other data studies. This research seeks to address the issue of over-exhaustiveness within K-Medoids algorithm by proposing a new method for selecting medoids which improves the efficiency of the created clustered and also reduce the number of iterations. Further, this research incorporates an objective to identify the best optimal value of k by using Silhouette Score which further improves the clustering accuracy and reliability. We employed different datasets to execute eight different algorithms, which included K-Means, K-Means++, Min-Max, K-Medoids, PAM, CLARA, CLARANS, and Optimized K-Medoid to quantifying clustering performance through metrics such as optimal clusters, iterations, and silhouette scores. The performance of the algorithms on clustering was assessed based on the optimal value of clusters; iterations performed, and silhouette scores. This proved that the Optimized K-Medoid offered the highest silhouette scores in the least iterations. Hence, this study shows that the Optimized K-Medoid Clustering algorithm is rather efficient than the other seven algorithms especially for clustering that require less time. In general due to its robust nature, it may be suitable for use in other fields of data analysis.