A modified scheme for all-pairs evolving fuzzy classifiers

Lughofer and Buchtala proposed the idea of all-pairs evolving fuzzy classifiers for multi-class classification. For each pair of classes, a binary classifier is used to classify all the training samples belonging to these classes. Two fuzzy classification architectures, singleton class labels and regression-based classifiers based on Takagi-Sugeno (T-S) models, are used as binary classifiers. The reference levels for pairs of classes are collected in the preference relation matrix. Finally, the preference relation matrix is used to determine the class to which the underlying input sample belongs. In this paper, we present a modified scheme for all-pairs evolving fuzzy classifiers. Two classifier architectures are proposed for binary classifiers. The first one combines the self-constructing fuzzy clustering (SFC) with the FLEXFIS-Class SM for singleton classifiers. The other one combines the SFC with the FLEXFIS-Class for regression-based classifiers. Experimental results demonstrate the effectiveness of the proposed modifications.