MR. FIS: Mamdani rule style fuzzy inference system
Damien Anderson, Lawrence Hall · 2003
Applying an adaptive fuzzy inference system to the input/output pairs produced by an artificial neural network will produce a set of rules that can be better understood by humans. The rules will model the artificial neural network providing a linguistic interpretation. These rules have triangular fuzzy sets in the antecedents and consequents to create what are often called Mamdani style rules. The resulting rules which model the performance of the artificial neural network will be meaningful and useful in explaining the operation of the artificial neural network. This paper presents MR. FIS, which stands for Mamdani rule style fuzzy inference system, a process to convert the knowledge contained in a neural network into Mamdani style fuzzy rules. Results on the well known Box-Jenkins dataset show the system effectively learns fuzzy rules. Results with fuzzy rules approximating learned neural networks are reported.