Using Curve Fitting in Error Correcting Output Codes

The Error Correcting Output Codes (ECOC) represent any number of the binary classifiers to model the multiclass problems successfully. In this paper, we have used Curve Fitting as a binary classifier in ECOC algorithm to solve multiclass classification problems. Curve Fitting is a classifier based on a nonlinear decision boundary that separates two pattern classes by the curves of the best fit, and arriving at optimal boundary points between two classes. Since we need a coding and a decoding strategy to design an ECOC system, this paper gives five coding and eight decoding strategies of ECOC and compares the results of Curve Fitting with Adaboost classification and Nearest Mean Classifier (NMC). This evaluation has been performed on different data sets of UCI machine learning repository. The results indicate that One-versus-one, ECOC-ONE coding and LAP, BDEN decoding having the best results in contrast with another coding and decoding strategies and Curve Fitting  is a good base classifier in ECOC, also it is comparable with the other ECOC approaches.

کلیدواژه ها:

classifier ، Coding ، Curve Fitting ، Decoding ، Error Correcting Output Codes (ECOC)