Generation of Fuzzy Classification Rules Directly from Overlapping Input Partitioning

The aim of this paper is to present a new method for extraction of fuzzy classification rules directly from numerical input - output data. The key feature of the proposed algorithm lies on the fact that it allows an overlapping between different classes. Appropriate membership functions are produced by projecting the geometrical characteristics of the corresponding classes on each input feature. The classification conflict is intuitively resolved by treating the overlapping regions separately, introducing double-consequent fuzzy rules. Finally, a fuzzy rule-based classification system is formalized, assembled, tested on Fisher Iris dataset and benchmarked against similar approaches.