Comparison of modified Karnik-Mendel algorithm-based interval type-2 ANFIS and type-1 ANFIS

Purpose This study aims to propose, as the first time, the interval type-2 adaptive network-fuzzy inference system (ANFIS) structure, which is given better results compared to previously presented in the open literature. So, the ANFIS can be used effectively for training of interval type-2 fuzzy logic system (IT2FLS) parameters. Design/methodology/approach Karnik-Mendel algorithm (KMA) is modified to use in interval type-2 ANFIS. The modified Karnik-Mendel algorithm (M-KMA) is implemented to change the uncertain ANFIS parameters into known ones. In this way, the interval type-2 ANFIS removes uncertainties of IT2FLS. Therefore, the interval type-2 ANFIS is reduced to a simple one, i.e. less mathematical operation required. Only consequent parameters are trained, and the consequent parameters are chosen in the form of crisp. Findings By applying the mentioned procedure, it can be shown that interval type-2 ANFIS has generally better results compared to type-1 ANFIS. However, it was noticed that the worst results obtained in the case of interval type-2 ANFIS are equal to the best result obtained in the case of type-1 ANFIS. Therefore, users in this field can use this approach in solving nonlinear problems. Practical implications The interval type-2 ANFIS can be used as controller for highly nonlinear systems such as air vehicles. Originality/value As stated in the open literature, it is ineffective to use ANFIS for IT2FLS. In this study, the KMA is modified for IT2FLS, and it is seen that the ANFIS can be used effectively for IT2FLS.

Automated summary

The study introduces the interval type-2 adaptive network-fuzzy inference system (ANFIS) structure for the first time, showing better results compared to previously presented methods. By modifying the Karnik-Mendel algorithm, uncertainties of ANFIS parameters are transformed into known ones, simplifying the system for effective training.