Global decision-making in multi-agent decision-making system with dynamically generated disjoint clusters

In this paper a new approach to organization of structure of multi-agent decision-making system, operating on the basis of dispersed knowledge, was proposed. In this approach dynamically generated disjoint clusters are used. The new approach is based on the assumption that one cluster should contain resource agents who agree on the classification of test object to the decision classes. This compatibility of agents is determined by certain relationships: a friendship relation and a conflict relation of agents. Definitions of friendship and conflict relations as well as method of determining the intensity of conflict between agents were taken from the paper of Zdzisław Pawlak [16]. In the process of global decision-making, the knowledge of resource agents remaining in such a coalition, is aggregated. On the basis of aggregated knowledge, local decisions, within clusters, are taken. In the last stage of global decision-making process a set of global decisions, using the method of conflict analysis, is generated. In this paper, the efficiency of inference of the multi-agent decision-making system with dynamically generated disjoint clusters were compared with the efficiency of inference of the system with statically defined clusters. For the overwhelming majority of the considered types of multi-agent decision-making system application of dynamically generated clusters have improved the efficiency of inference. The best results were obtained for data sets: Dermatology, Vehicle Silhouettes and Landsat Satellite, slightly worse results were obtained for the Soybean Data Set. It should also be noted that the greater the number of resource agents in multi-agent decision-making system, the greater the expected improvement in the efficiency of inference, when applied dynamically generated clusters.