Ternary Bradley-Terry model-based decoding for multi-class classification

A multi-class classifier based on the Bradley-Terry model predicts the multi-class label of an input by combining the outputs from multiple binary classifiers, where the combination should be a priori designed as a code word matrix. According to this framework, the code word matrix was originally designed to consist of +1 and -1, and has later been extended to allow zero components. This extension has seemed to effectively work, but in fact, contains a problem. In this article, we propose a Boosting algorithm, which deals with three categories by allowing a dasiadonpsilat carepsila category, and present a modified decoding method called dasiaternarypsila Bradley-Terry model. In addition, we propose a fast decoding scheme which resolves the heavy computation of the conventional Bradley-Terry model-based decoding.