Performance Evaluation of Some Clustering Algorithms under Different Validity Indices
Samira Faisal Abushilah, Rajaa Hasan Abbas · Mathematical Modelling and Engineering Problems · 2023
Clustering, a pivotal technique in statistics, enables the summarisation of data sets through the identification of related object groups.A prevalent question in clustering literature pertains to the precise number of partitions present within a data set.An array of clustering methods and indices has been proposed to discern the optimal number of clusters within a data set, each following its own set of rules.However, none of these methods universally excel in capturing the true components across all types of data structures.Particularly, they tend to grapple with uniquely shaped data sets or instances where objects from different groups are in close proximity.In this study, the performance of several clustering methods (Single Linkage, Complete Linkage, Average Linkage, Centroid Linkage, Ward.2DLinkage, Median Linkage) is evaluated in conjunction with different internal validity indices (KL, CH, Sil, Gap).This evaluation utilises simulated data, encompassing varied models, sample sizes, and distance measures, and is conducted using R software 3.1.Furthermore, several external indices (Rand, F-M, Purity) are employed to ascertain the degree of agreement between the true clusters of data points and the partitions computed through the clustering methods.