A Multi-Agent Classifier System based on Fuzzy-ARTMAP and Fuzzy Min-Max Neural Networks

Abstract

In this paper, we propose a Multi-Agent Classifier (MAC) system based on the Trust, Negotiation, and Communication (TNC) model. A novel trust measurement method, based on the recognition and rejection rates, is proposed. Two agent teams, each consists of three neural network (NN) agents, are formed. The first is the Fuzzy Min-Max (FMM) NN agent team and the second is the Fuzzy ARTMAP (FAM) NN agent team. An auctioning method is also used for the negotiation phase. The effectiveness of the proposed model and the bond (based on trust) is measured using two benchmark classification problems. The bootstrap method is applied to quantify the classification accuracy rates statistically. The results demonstrate that the MAC system is able to improve the performances of individual agents as well as the team agents. The results also compare favorably with those from other methods published in the literature.