Recursive classifiers

A recursive approach for the design of non-binary classifiers is proposed. By means of recursive coding models and the machine learning error-correcting output codes (ECOC) framework, learning in nonbinary output domains is reduced to a set of binary learning problems and a combination algorithm. This recursive learning formulation, hereafter named as RECOC learning, allows the design of general error adaptive classifiers, thus generalizing the binary-boosting concept.