On Using Structural Patterns In Data For Classification

Abstract

There are some interesting approaches for classification such as semi-supervised algorithms, algorithms that learn distance functions, and various extensions and generalizations of support vector machines. In this study, we propose a new clustering algorithm that uses similarities only and is used as an intermediate step for classification. The motivation for this combined approach is to obtain information from the data set that can be used for classification. After obtaining a clustering of the data set with the proposed clustering algorithm, we apply different strategies for classification. The results on some data sets show that this approach can have some advantages. For example, when using support vector machines, the size of the training set is reduced, while at the same time, comparable performance results are obtained with a smaller number of support vectors.