Facing Polychotomies through Classification by Decomposition: Applications in the Bio-medical Domain

Abstract

Polychotomies are recognition tasks with a number of categories greater than two, consisting in assigning patterns to a finite set of classes. Although many of the learning algorithms developed so far are capable of handling polychotomies, most of them were designed by nature for dichotomies, that is, for binary learning. Therefore, various methods that decompose the multiclass recognition task in a set of binary learning problems have been proposed in the literature. After addressing the different dichotomies, the final decision is reconstructed according to a given criterion. Among the decomposition approaches, one of them is based on a pool of binary modules, where each one distinguishes the elements of one class from those of the others. For this reason, it is also known as one-per-class method. Under this decomposition scheme, we propose a novel reconstruction criterion to set the final decision on the basis of the single binary classifications. It looks at the quality of the current input and, more specifically, it is a function of the reliability of each classification act provided by the binary modules. The approach has been tested on six biological and medical datasets (two private, four public) and the achieved performance has been compared with the one previously reported in the literature, showing that the method improves the accuracies achieved so far.