Optimizing ECOC matrixes in Multi-Class Classification Problems
Computer Science and Engineering, MSc Thesis, 2013
Assoc Prof. Berrin Yanıkoğlu (Thesis Supervisor), Asst. Prof. Hakan Erdoğan, Assoc. Prof. Gözde Ünal, Assoc. Prof. Nilay Noyan, Asst. Prof. N.Özben Önhon
Date &Time: August, 05th, 2013 – 15:30
Place: FENS L055
Keywords: ECOC, error correcting output codes, ensemble learning, multi-class classification
Error Correcting Output Coding (ECOC) is a multi-class classification technique in which multiple binary classifiers are trained according to a preset code matrix, such that each one learns a separate dichotomy of the classes. While ECOC is one of the best solutions to multi-class problems, it is suboptimal since the code matrix and the base classifiers are not learned simultaneously. In this thesis, we present three different algorithms that iteratively updates the ECOC code matrix to improve the performance of the ensemble by reducing the decoupling. Firstly, we applied FlipECOC+ method which first finds all matrix entries below specified value then flips them according to ascending order of their accuracy. It accepts the updated ECOC matrix if the validation accuracy improves. Second method is appyling simulated annealing method on updating ECOC matrix by flipping proposed entries according to ascending order. It accepts the updated ECOC matrix according to probability. Last method is applying beam search to find updated ECOC matrix which has highest validation accuracy. We applied all three algorithms on UCI (University of California Irvine) data sets. Beam search algorithm gives the best result on UCI data sets.All of the proposed update algorithms does not involve further training of the classifiers and can be applied to any ECOC ensemble.