A dispersed decision-making system – The use of negotiations during the dynamic generation of a system’s structure

• Two-stage process of creating coalitions in a dispersed decision system was applied.
• Three types of relationships were used – friendship, conflict and neutrality.
• Initial clusters are formed in the first stage – groups of classifiers in friendship.
• In the second stage (a negotiation stage) neutral classifiers play a crucial role.
• New approach of generating clusters has been compared to approach proposed in [22].

The issues that are related to the process of global decision-making on the basis of knowledge which is stored in a dispersed form (several local knowledge bases or classifiers) are discussed in this paper. In a decision-making system, which is described in the paper, the classification process of the test object starts with an investigation of how particular classifiers classify a test object. We describe the views of classifiers by using probability vectors over decision classes. In the system, the process of combining classifiers in coalitions is very important. Negotiation is used in the clustering process. We define three types of relations between classifiers: friendship, conflict and neutrality. The clustering process consists of two stages. In the first step, the initial groups are created. These groups contain classifiers that are in a friendship relation. In the second stage, classifiers which are in neutrality relation are attached to the existing groups. In this paper, a formal description of the clustering process is presented and mathematical properties of functions, which are used, are described. For every cluster, we find a kind of combined information. Finally, we classify the given test object by voting among clusters, using the combined information from each of the clusters.In the paper a new way of creating clusters (with a negotiation stage) is compared to the approach presented in the paper (Przybyła-Kasperek and Wakulicz-Deja, 2014) [23] (without negotiations). There are significant differences between the clusters that are generated using these two approaches, which are shown in the paper. In the new approach, the clusters are more complex and better reconstruct and illustrate the views of the classifiers on the classification.