Optimization of fuzzy procedure for classification of medical problems

The author has modified Kosko´s standard additive model (1997) of fuzzy systems to handle classification problems. This involves overcoming the curse of dimensionality by limiting the number of rules to the number of classes. The basic fuzzy classification procedure (FCP) does not use weights. Using weights on the fuzzy rules should improve the FCP´s accuracy. Investigating how to choose weights that improve accuracy without lengthy training is the purpose of the research reported in this paper. If the weights smoothed the accuracy for each class, as well as improving overall accuracy, this would be an added benefit, especially for medical problems The basic FCP uses the vector of means for each class as the class prototype. The fuzzy rules can be applied to these vectors to produce a matrix of scores. The greater the difference between the diagonal elements and the other elements on the same row the better FCP should perform. These weights improved the performance of the basic fuzzy procedure from 2-20% for several problems, mainly medical. It not only improved the overall accuracy of prediction; it “smoothed” the accuracy, so that, in the medical problems, the two classes were each predicted with comparable accuracy