A Comparison of Error Correcting Output Coding Methods for Multiclass Classification by Using Support Vector Machine: The Prediction of Self-Monitoring of Blood Sugar

Abstract

SVMs were initially developed to perform binary classification. However, in many real-world problems, particularly pattern recocnation studies, aimed to determine the distinctive features of large number of class or group. For this reason, a number of methods to generate multiclass SVMs from binary SVMs have been proposed by researchers and this is still a continuing research topic. In this study we aimed to compare classification accuracy and computational cost of four multiclass approaches using a original and simulated data sets. Error Corrected Output Coding (ECOC) based multiclass approaches that is used in this study creates many binary classifiers and combines their results to determine the class label of a test pixel. As a result of the comparisons for all conditions examined in this study, it’s found that the classification accuracy and computational cost of One vs. One multiclass approach is better than the other multiclass approaches. In classification or pattern recognization problems, some of supervised machine learning methods or algorithms can be easily extended to multiclass problems. However, some other powerful and popular classifiers, such as AdaBoost and Support Vector Machines, do not extend to multiclass easily. In those situations, the usual way to proceed is to reduce the complexity of the multiclass problem into multiple simpler binary classification problems.