Attributes Regrouping in Fuzzy Rule Based Classification Systems: An Intra-Classes Approach

Fuzzy rule-based classification systems (FRBCS) are able to build linguistic interpretable models, they automatically generate fuzzy if-then rules and use them to classify new observations. However, in these supervised learning systems, a high number of predictive attributes leads to an exponential increase of the number of generated rules. Moreover the antecedent conditions of the obtained rules are very large since they contain all the attributes that describe the examples. Therefore the accuracy of these systems as well as their interpretability degraded. To address this problem, we propose to use ensemble methods for FRBCS where the decisions of different classifiers are combined in order to form the final classification model. We are interested in particular in ensemble methods which split the attributes into subgroups and treat each subgroup separately. We propose to regroup attributes by correlation search among the training set elements that belongs to the same class, such an intra-classes correlation search allows to characterize each class separately. Several experiences were carried out on various data. The results show a reduction in the number of rules and of antecedents without altering accuracy, on the contrary classification rates are even improved.