Empirical Study on Error Correcting Output Code Based on Multiclass Classification

A common way to address a multi-class classification problem is to design a model that consists of hand picked binary classifiers and to combine them so as to solve the problem. Error-Correcting Output Codes (ECOC) is one such framework that deals with multi-class classification problems. Recent works in the ECOC domain has shown promising results demonstrating improved performance. Therefore, ECOC framework is a powerful tool to deal with multi-class classification problems. The error correcting ability improve and enhance the generalization ability of the base classifiers. This paper introduces state-of-the-art coding (one-versus-one, one-versus-all, dense random, sparse random, DECOC, forest-ECOC, and ECOC-ONE) and decoding designs (hamming, Euclidean, inverse hamming, laplacian, β-density, attenuated, loss-based, probabilistic kernel-based, and loss weighted) perspectives along with empirical study of ECOC following comparison of various ECOC methods in the above context. Towards the end, our paper consolidates details relating to comparison of various classification methods with Error Correcting Output Code method available in weka, after carrying out experiments with weka tool as a final supplement to our studies.