A study on differential evolution-optimized ANFIS

Abstract

ANFIS rely on gradient-based optimization methods, which are prone to local optima, slow convergence, and sensitivity to parameter initialization. To address these limitations, this study proposes a Differential Evolution–optimized ANFIS framework. The proposed approach integrates FCM clustering to initialize the fuzzy rule base, thereby improving structural interpretability and reducing initialization uncertainty. Subsequently, DE is employed to globally optimize both antecedent and consequent parameters of ANFIS, enhancing its global search capability and convergence performance. The optimal number of fuzzy rules is determined as R=2 using cluster validity analysis based on PC and XB index. In addition, DE parameters are set as mutation factor F=0.30 and crossover rate CR=0.50 according to sensitivity analysis. Experimental evaluation is conducted on an apple leaf disease dataset using 5-fold cross-validation. The proposed DE-optimized ANFIS framework achieves approximately 90% classification accuracy, demonstrating superior robustness, convergence stability, and generalization ability compared to conventional ANFIS models.