A novel approach to classificatory problem using neuro-fuzzy architecture

Abstract

Classification is one of the major problems solved by the artificial neural networks. The problem deals with the mapping of the input data into classes. Voice recognition, pattern matching, face recognition, character recognition etc. are these types of problems. In the past few years we have seen a great increase in the methods to solve these problems. In this paper we have proposed a new method for solving these problems. The method is inspired from the neuro-fuzzy logic approach to problem solving. Here we have proposed the various changes that may be made in various layers for the system to better handle classificatory problems. We first cluster the training data so that the rest of the algorithm works efficiently with limited data to handle. The clustering is based on class identification of various inputs given as training data. A sort of fuzzy logic approach serves as a means to classify the unknown input to a class based on the rules formed from inputs given in training data. Rules are in form of representative of every cluster and their matching class. The centre and power of the representative are the parameters that are optimized using a training algorithm similar to the backpropagation algorithm of the artificial neural networks. This algorithm is optimized by Genetic Algorithms. We tested the algorithm on the famous classificatory problem of picture learning. We got a good efficiency which proves the effectiveness of the algorithm.

Recurrent Neuro-Fuzzy Control of Grid-Interfaced Solid Oxide Fuel Cell (SOFC) System

“…Since both fuzzy logic and ANN had their relative benefits, a powerful processing with benefits of both was achieved by merging them together. The learning ability of Neural Network was used for regulating the parameters of fuzzy logic in various scenarios for achieving better performance (Chu and Tsai, 2008;Afzalian and Linkens, 2000;Kala et al, 2011;Ananthababu et al, 2016;Khan et al, 2016;Sardar et al, 2015). Nevertheless, application field was restricted to static problems as the feed forward network composition was a major shortcoming of the neuro-fuzzy system.…”

Classification using redundant mapping in modular neural networks

“…Another approach makes use of an algorithm inspired by neuro-fuzzy systems for better classification [11,12]. Here the training inputs are first clustered as per their classes, and then individual cluster representatives are used for problem solving.…”