A NOVEL FUZZY INFERENCE MODEL WITH RULE-BASED DEFUZZIFICATION APPROACH

Nasiboğlu R.

Journal of Modern Technology and Engineering, cilt.7, sa.2, ss.124-133, 2022 (Hakemli Dergi)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 7 Sayı: 2
  • Basım Tarihi: 2022
  • Dergi Adı: Journal of Modern Technology and Engineering
  • Derginin Tarandığı İndeksler: Other Indexes
  • Sayfa Sayıları: ss.124-133
  • Dokuz Eylül Üniversitesi Adresli: Evet

Özet

Fuzzy inference system (FIS) is one of the most used approaches in decision making and machine learning models. The main advantage of this system is that it can process data in the fuzzy form whose exact value is unknown. Various FIS models are available in the literature. The most common FIS models are Mamdani and Sugeno type models. In Sugeno type FISs, the output of each rule should be given as a specific mathematical function. Sometimes these functions can be difficult to identify. In the Mamdani model, it is sufficient to give an approximate fuzzy set instead of the specific function as a rule output. The overall output of the Mamdani type model is the aggregated fuzzy set, which is an aggregation of individual rule outputs. But in this composite set, the contours of each rule output are lost and the effect on the overall output is not very obvious. A novel type FIS with rule-based defuzzification (FIS-RBD) is proposed in this study. This model is a Mamdani type FIS, however, the defuzzification process is applied to each rule’s output instead of the overall output. The overall output of the system is calculated as the weighted average of the defuzzified values of the rule outputs. This approach is a synthesis of a Mamdani type and Sugeno type models. In the study, the details of the proposed approach are examined and the working principle is explained with numerical examples.