A Data Labeling Method for Categorical Data Clustering Using Cluster Entropies in Rough Sets

Data mining automates the process of finding predictive records in large databases. Clustering is a very popular technique in data mining and is a significant methodology that is performed based on the principle of diminishing the intra-cluster dissimilarities and in turn increases the inter-cluster dissimilarity. The segregation of a large database is a challenging and time consuming task. For this purpose, an approach called data labeling through sampling technique is used. Using this approach segregating large databases not only gets easier but also it increases the efficiency of clustering technique. Initially a sample data is retrieved from a large database for clustering and the residual unsampled data points are compared with the clustered data from which the similar data points are clustered and the dissimilar one are considered as outliers based on various data labeling techniques. These data labeling techniques are easier to apply in the numerical domains, whereas in the categorical domains this is a complicated task as the distance among data points are incalculable. Further the proposed methodology gives a data labeling technique based on the changes in the intra-cluster and inter-cluster dissimilarities after including unlabelled data point into existing cluster for categorical data using cluster entropy in rough set theory. The experimental results show that the proposed algorithm is an efficient and high quality clustering algorithm compared to that of the previous ones.