Efficiency analysis of the rule-based defuzzification approach to fuzzy inference system for regression problems

A fuzzy inference system (FIS) is an effective prediction method based on fuzzy logic.The performance of this model may vary depending on the defuzzification process.In the Mamdanitype FIS model, the defuzzification process is applied to the fuzzy output of the system only once at the last stage.In the FIS with rule-based defuzzification (FIS-RBD) model, the defuzzification process is applied to the fuzzy consequent part of each rule and the overall result of the system is calculated as the weighted average of the separately defuzzified results of the rules.Note that, the original shapes of the combined rule results are lost in the aggregated fuzzy result of the classical Mamdani-type system and the effect of each rule on the system result decreases when aggregated.However, rule results can affect the overall result more significantly in the FIS-RBD approach.In this study, a comparative analysis was made on the effectiveness of the classical Mamdanitype FIS and FIS-RBD models for regression problems.Five datasets from different domains and various defuzzification methods were used in comparisons.In the results obtained, it was observed that the The FIS-RBD model gave better results than the classical Mamdani-type FIS model.To carry out calculation experiments, a new Python package called Fuzlab was developed by modifying the existing Python library called FuzzyLab.In addition to creating the FIS-RBD model, the developed package also allows the use of the Weighted Average Based on Levels (WABL) defuzzification method in fuzzy logic-based calculations.