FUZZY CLASSIFIER TRAINING WITH ONLY MAIN COMPETITORS

Abstract

The tie "input—output" is described by linguistic «if—then» rules where antecedents contain fuzzy terms "low", "medium", "high" in the fuzzy classifiers. To enhance the correctness it is necessary to train fuzzy classifier on experimental data. There have been proposed new criteria for fuzzy classifier training that take into account the difference of fuzzy output only to the main competitors. When the classification is correct the main competitor of the decision is the class with the second largest degree of membership. In cases of misclassification erroneous decision is the main competitor to the correct class.

Computer experiments with the tuning up of a fuzzy classifier for UCI-problem of recognition of Italian wines showed a significant advantage of the new training criteria. New criteria of training can be used not only for tuning fuzzy classifiers but for some other models, such as neural networks.