A subspace approach to error correcting output codes

Among the proposed methods to deal with multi-class classification problems, the error-correcting output codes (ECOCs) represents a powerful framework. A key factor in designing any ECOC matrix is the independency of the binary classifiers, without which the ECOC method would be ineffective. This paper proposes an efficient new approach to the classical ECOC design in order to improve independency among classifiers. The main idea of the proposed method is based on using different feature subsets for each binary classifier, named subspace ECOC. In addition to creating more independent classifiers in the proposed technique, ECOC matrices with longer codes can be built. The numerical experiments in this study compare the classification accuracy of subspace ECOC, classical ECOC, one-versus-one, and one-versus-all methods over a set of UCI machine learning repository datasets and two image vision applications. The results show that the proposed technique increases the classification accuracy in comparison with the state of the art coding methods.

► We propose a new approach to ECOC by using different feature sets for each classifier.
► In this way, we generate three-dimensional codematrix.
► The method improves independency among binary dichotomizers.
► Classification accuracy improves in comparison with the state-of-the-art methods.