Neo-Fuzzy Supported Brain Emotional Learning Based Pattern Recognizer for Classification Problems

Abstract

Based on the limbic system theory of mammalian emotional brain, supervised brain emotional learning-based pattern recognizer (BELPR) has been recently proposed for multi-input and multi-output classification problems. It offers features like: decreased time and spatial complexity, faster training and higher accuracy. BELPR has been deployed to classify a number of benchmark datasets and has demonstrated its superior performance compared with the conventional multilayer perceptron network. The goal of this paper is to further enhance the classification accuracy of BELPR through integration with Neo-Fuzzy Neurons (NFN). The network built using NFN shares many of the same characteristics as BELPR, such as: simplicity, transparency, accuracy, and lower computational complexity. With this view in mind, this paper proposes a new neuro-fuzzy hybrid classification network: Neo-Fuzzy supported brain emotional learning-based pattern recognizer (NFBELPR), which will preserve the features of both networks, while simultaneously improving the performance of BELPR. The NFBELPR model can be considered as a group of two networks depending upon the level of integration of NFN and BELPR. When the integration of NFN is only considered in the orbitofrontal cortex section of BELPR, the resulting classification model is termed as partially integrated NFBELPR. In cases, when the integration is considered both in the OFC and amygdala sections of BELPR, the resulting classification model becomes fully integrated NFBELPR. The proposed NFBELPR networks are implemented in MATLAB®R2009b programming environment to classify a number of benchmark datasets. They are found to achieve higher classification accuracy when compared with BELPR and some state of the art classification networks.