Nonlinear modelling and classification based on the MICOG defuzzification Original Research Article

In the paper we study the MICOG (modified indexed center of gravity) defuzzification applied to neuro-fuzzy architectures with various inference methods. It is proven, that the MICOG defuzzification can be used in a combination with the Mamdani-type reasoning and selected S and Q fuzzy implications. It is also shown that the MICOG defuzzification cannot be used with R fuzzy implications if we require interpretable rules. The experiments confirm our theoretical results.