A Modified Initialization Method to Find an Initial Center for Fuzzy K-Modes Clustering

Data mining is used to analyzing the data and extracts the information from large data sets. Clustering is a technique which is used to grouping the objects which is having similar characters. In existing system, K-Means clustering is used to cluster the numerical data but it does not cluster the categorical data. To overcome the limitation of using numeric data only in the clustering process, F Jiang proposed the initialization method for K-modes clustering in which the initial cluster centres were calculated using Ini_Distance and Ini_Entropy methods were used. To improve the performance further, a new modified model is proposed for Fuzzy K-modes clustering. In this model, a modified Initial_Distance and Initial_Entropy Initialization method are used. It ensures that K initial cluster centers are not selected from the same cluster and selected initial cluster centers are not outliers. Also, one of a data pre-processing methods-Data cleaning is used to increase the quality of clustering results and a new modified weight matching distance (WMD) is calculated by using distance between two objects at the time initialization. An experiment result is demonstrated with five data sets taken from UCI machine learning repository and the results proved the effectiveness of the initialization method.