Multi-label Testing for CO2RBFN: A First Approach to the Problem Transformation Methodology for Multi-label Classification

Abstract

While in traditional classification an instance of the data set is only associated with one class, in multi-label classification this instance can be associated with more than one class or label. Examples of applications in this growing area are text categorization, functional genomics and association of semantic information to audio or video content. One way to address these applications is the Problem Transformation methodology that transforms the multi-label problem into one single-label classification problem, in order to apply traditional classification methods. The aim of this contribution is to test the performance of CO2RBFN, a cooperative-competitive evolutionary model for the design of RBFNs, in a multi-label environment, using the problem transformation methodology. The results obtained by CO2RBFN, and by other classical data mining methods, show that no algorithm outperforms the other on all the data.