A framework for classification using genetic algorithm based clustering

Abstract

Clustering has been used in literature to enhance classification accuracy. But most partitional clustering methods need the number of clusters as input and also they are sensitive to initialization. Although hierarchical clustering methods may be more effective in finding clustering structure of the dataset than partitional methods but hierarchical clustering methods give tree structure known as dendrogram which is a sequence of clustering solutions. Hence hierarchical clustering algorithms are not generally applied in the preprocessing step to classification methods. This problem can be solved by cutting the dendrogram to get single clustering solution. In this paper we propose a framework for classification which uses Optimal Clustering Genetic Algorithm (OCGA) to obtain optimal level of cutting the dendrogram. A single clustering solution is obtained by cutting the dendrogram at optimal level. The clusters obtained are used to enhance classification accuracy of the classification methods. The proposed classification methods have been applied for the diagnosis of diabetes disease.