A Novel Hybrid Clustering System using k-means, k-medoids, hierarchical, Fuzzy C Means Algorithms on Thyroid Drug Data using R

Katikireddy Srinivas, Dr K V D Kiran

Katikireddy Srinivas, Dr K V D Kiran

Abstract

Bunching calculations section a dataset into different gatherings or groups by disseminating the items in a specific gathering that have a high level of likeness and the articles in various gatherings show greater difference. About 189 medications which are known to act against thyroid infection with 33 huge property estimations are removed from MalaCards database was chosen as a dataset. Properties of medications assessed from Drug Bank database. So as to accomplish better outcomes in bunch examination, Hopkins measurement was utilized to survey the grouping inclination of a dataset.
An estimation of 0.3313 was gotten which recommends that the information is consistently disseminated and exceptionally clusterable. Further, elbow and outline procedures and NbClust bundle utilized to decide ideal groups which brought about 3 bunch arrangements as ideal. With k=3, the k-means and k-medoids calculation brought about 3 bunches of different sizes, while half and half k-implies k-medoids approach brought about predominant groups sizes. Utilizing the half and half methodology diminished the quantity of negative outlines in examination.
We present a cross breed technique with both algorithms;k-medoids and k-intends to bunch a dataset of thyroid illness drugs and the program is hurried to create groups focused on k-means and k-medoids, trailed by improving the result by executing fluffy k-means. Clusterability was completed by Hopkins measurement and group legitimacy by Nbclust came about in k=3. Both the techniques brought about groups with negative outlines, in any case, half breed bunching calculation brought about incomplete covering of information focuses, henceforth fluffy k-implies calculation was applied on sub-set of dataset. At last, of all the six fluffy calculations contemplated, fkm calculation showed prevalent partition of bunches with all around characterized information focuses.