Polychotomous Classification with Pairwise Classifiers: A New Voting Principle

Abstract

A new principle for performing polychotomous classification with pairwise classifiers is introduced: if pairwise classifier $\mathcal{N}_{ij}$, trained to discriminate between classes $i$ and $j$, responds ``i'' for an input $\myvec x$ from an unknown class (not necessarily $i$ or $j$), one can at best conclude that $\myvec{x} \not \in j$. Thus, the output of pairwise classifier $\mathcal{N}_{ij}$ can be interpreted as a vote against the losing class $j$, and not, as existing methods propose, as a vote for the winning class $i$. Both a discrete and a continuous classification model derived from this principle are introduced.