Comparison of Hard and Fuzzy Clustering Techniques and Selection of Optimal Fuzzifier Parameter: An Application on Household Characteristics and Health Expenditures

Authors : Songül Çinaroğlu

Pages : 17-31

View : 69 | Download : 80

Publication Date : 2024-01-08

Article Type : Research Paper

Abstract :It is a challenging task for decision makers for finding the optimal classification pattern for the dataset obtained from national accounts, such as household budget survey (HBS) data. Fuzzy c-means (FCM) clustering, a fuzzy logic-based clustering algorithm, can be used effectively to find the proper cluster structure of given data sets under uncertainty. In this study, crisp (k-means) and fuzzy (FCM) clustering performances on grouping of households are compared while changing fuzzifier parameter for FCM. The results of the study reveal that FCM clustering performs better when compared with k-means clustering. It is found out that the optimal number of household groups is 5 and further, high cluster validity index scores are obtained when fuzzifier value is 1.5 in FCM clustering. High cluster validity index scores obtained from fuzzy Silhouette is compared to the crisp cluster validity index. The experimental results proved that fuzzy clustering superior grouping ability and it has better validity measures for grouping of households in a national dataset. It is observed that smaller fuzzifier value is a better choice to enhance fitness of fuzzy clustering. It is hoped that future experiments will compare the clustering abilities of FCM using datasets with different sizes and variables under the uncertainty conditions to determine the class boundary.
Keywords : fuzzycmeans, fuzzifier parametresi, kmeans, silhouette