Comparison between error correcting output codes and fuzzy support vector machines

Abstract

One-against-all support vector machines with discrete decision functions have unclassifiable regions. To resolve unclassifiable regions, support vector machines with continuous decision functions and fuzzy support vector machines have been proposed. If, in ECOC (error correcting output code) support vector machines, instead of discrete error functions, continuous error functions are used, unclassifiable regions are resolved. In this paper, first we prove that for one-against-all formulation, support vector machines with continuous decision functions are equivalent to fuzzy support vector machines with minimum and average operators. Then we discuss minimum operations as well as average operations for error functions of support vector machines and show the equivalence of ECOC support vector machines and fuzzy support vector machines for one-against-all formulation. Finally, we show by computer simulations that ECOC support vector machines are not always superior to one-against-all fuzzy support vector machines.