The Introduction of Gini`s Mean Difference in Weighted Euclidean Distance Clustering Analysis

Abstract

The cluster analysis has newly become extremely vigorous issue in data mining reaearch. The main aim is to disintegrate a dataset into dissimilar subset clusters or groups, whereby, data in a particular subset have the same membership and characteristics while different subset presenting different membership from data in different subset. We introduced Gini`s mean difference (GMD) to replace standard deviation in Standardized Euclidean distance based on some simulation study results experimented by some scholars in the literature was found to be less effective in many circumstances. However, the GMD is superior to standard deviation when the measure of variability is non-normal distributions and effective for clustering overall distribution is composed to subsets. The suggested method is called Gini`s mean difference weighted Euclidean distance. To examine the performance of the suggested method simulation study and real data applications are considered. Consequently, evidently, the suggested method has shown good performance compared to the existing methods, by achieving nearly maximum point in average external validity measures, recorded lower computing time and clustered the sample size to their maximum prior assigned clusters. However, from the results obtained, it could be said the proposed method achieved much better performance compared to the existing methods based on the simulation study and real data applications experimented on cluster analysis.