Use of Fuzzy Clustering Algorithms Ensemble for SVM Classifier Development

Abstract

This paper represents an approach for object's classification based on the combined implementation of the SVM algorithm and the fuzzy clustering algorithms. It is offered to use the fuzzy clustering algorithms’ ensemble based on the clusters’ tags’ vectors’ similarity matrixes and the spectral factorization algorithm as for creation the training and test sets, used for development of SVM classifier, as for specification of the clustering results with application of the trained SVM classifier. In the first case we solve an absence problem of aprioristic information on objects’ possible class association and avoid inclusion of noise objects in the training and test sets, that will provide development of more exact SVM classifier. In the second case we use part of the clustering results received by means of the fuzzy clustering algorithms’ ensemble for training of SVM classifier, and another – for the specification of the clustering results by means of the trained SVM classifier. We use in the spectral factorization algorithm along with the k-means algorithm, which is usually applied to graph splitting into clusters, the FCM algorithm, allowing to reduce the cost of graph splitting and receive better graph splitting into clusters, that will provide higher quality of objects’ classification.