Design and Development of An Auto Adaptive Multi-Vigilance for Simplified Fuzzy ARTMAP

Abstract

This paper presents the design and development of a Simplified Fuzzy Resonance Theory (SFAM) with an auto-adaptive multivigilance parameter. Using the concept of adaptation from human familiarity in recognizing things to make a decision of patterns of the artificial neural network more flexible and efficient. Based on the original SFAM architecture that has only one vigilance parameter is defined and it is not adaptive. This vigilance parameter was designed to compare the similarity of the dataset to winner neurons stored in the neural weight layer. causing the flexibility in the model of the artificial neural network to decrease Therefore, an idea was born to design and develop the architecture of a simple adaptive fuzzy resonance theory network to be more flexible by designing an algorithm that can generate multiple vigilance parameters based on the number of nervous weight neural and this parameter can be adapted according to the learning process that the neural network receives from the dataset. Two types of datasets were tested including 1. simple input dataset and 2. complex input dataset. The test results show that the efficiency of the purpose neural network has an accuracy percentage of 96.67% and 95.33% respectively. While traditional networks have accuracy percentages of 86.67% and 92.66%.