A meta-fuzzy classifier for specifying appropriate fuzzy partitions by genetic fuzzy rule selection with data complexity measures

Tens of thousands of classifiers have been proposed so far. There is no best classifier among them for all the existing data sets. The performance of each classifier often depends on the data sets used for comparison. Even for a single classifier, suitable parameters of the classifier also depend on the data sets. That is, there is a possibility that a suited classifier and its parameter specification can be chosen beforehand if the target data sets or their characteristics were known. In recent years, a number of data complexity measures have been proposed to characterize data sets. The aim of this study is to develop a meta classifier for selecting an appropriate classifier and/or its appropriate parameter specification by means of data complexity measures. In this paper, we focus on the parameter specification of fuzzy classifiers using data complexity measures as a preliminary study. To construct a meta-classifier, we generate a large number of artificial data sets from Keel benchmark data sets. Then we generate meta-patterns which are composed of the values of data complexity measures as inputs and an appropriate fuzzy partition as an output. Using meta-patterns, a meta classifier is designed by multiobjective genetic fuzzy rule selection. We evaluate the proposed method through leave one-group out cross-validation.