FLIP-ECOC: A Greedy Optimization of the ECOC Matrix
Error Correcting Output Coding (ECOC) is a classication technique designed for multiclass classication problems. In this approach, multiple dichotomizers are trained using subsets of the training data, determined by a preset code matrix. While ECOC is one of the best solutions to multiclass problems, the solution is suboptimal due to the fact that the code matrix and the dichotomizers are not learned at the same time. In this paper, we show an iterative update algorithm for the code matrix that is designed to reduce this decoupling. We compare the proposed algorithm with the basic ECOC approach for dierent number of dichotomizers and show that it improves the base ECOC accuracy, for some well-known data sets.