Error correcting output coding

This is a method discussed in paper Ditterich and Bakiri, JAIR 1995.

Error Correcting Output Coding (ECOC) converts a m-class supervised learning problem into a n binary problems. The algorithm assigns each class a unique binary codewords of length n. Then it trains n classifiers to predict each bit in the codewords. The predicted class is the one whose codewords is the closest to the one predicted by the classifier.