A study comparing performances used algorithms in categorical data analysis

Cluster analysis is a method used to find natural groups of objects. Given a data set the main goal is to produce a partition with high internal intra-cluster similarity and high inter-cluster dissimilaity. Clustering with numerical data is quite easy and there are many methods for them. But clustering of categorical data is more difficult than clustering of numerical data. There is not many methods for clustering of categorical data and there is no certain information about which one is best. According to the number of data and data sutructure each has advantages and limitations. Also variable number is important for good clustering results. In this thesis dealt with clustering of categorical data. Hierarchical clustering techniques which are single linkage, complete linkage, average linkage and partitional clustering technique which is K-modes algorithm were compared. Well known real data sets were used for quality comparison. According to the analysis results when the number of data set grows clustering performances are decreasing in single linkage, complete linkage, average linkage while K-modes algoritm?s is increasing.