Comparison of the Effects of Different Dimensional Reduction Algorithms on the Training Performance of Anfis (Adaptive Neuro-Fuzzy Inference System) Model

Abstract

Adaptive Neuro-Fuzzy Inference Systems (ANFIS) is a hybrid artificial neural network (intelligence) approach that utilizes the ability of artificial neural networks to learn, generalize, paralyze and to derive fuzzy logic. The development of models with large numbers of input variables with ANFIS is not very convenient for applications. Dimension reduction methods are proposed as a solution to this problem. Dimensional Reduction is the method used to represent the data in a lower dimensional space. The reduction of the numbers of the input variables using different size reduction methods and the creation of the optimal solution of the probing with the ANFIS model constitute the framework of this work. In this study, we compared the results produced by different dimension reduction methods and investigated which method is more acceptable for ANFIS training.

Feature Space Reduction, using PCA in the Algorithm for Epilepsy Detection, using an Adaptive Neuro-Fuzzy Inference System and Comparative Analysis

“…The comparisons of the accuracies of the old and the upgraded algorithm for the different data split methods and the different input space partitioning were made both for 40 and 60 epochs, as with the old algorithm, in some of the cases, overfits after 40 epochs and 60 epochs seemed a reasonable sublimate according the RMSE simulation results. It is well known that size reduction methods are preferred, in applications where the number of input parameters in the data set is too large [31]. In this sense, the PCA in the upgraded algorithm plays a key role, i.e.…”