A first approximation in order to define a Difficulty Factor of the bi-classification in a dataset by using SVMs

Abstract

The main aim in this paper is to analyze the complexity of a Support Vector Machine -SVM- in the construction of a classifier for a bi-classification problem on a specific dataset. Hence, an index is defined in terms of both, the Lagrange multipliers and the number of support vectors. Experimentation for cheching the defined index is carried out with a well.known dataset, the Glass Identification Database.