Global decision-making system with dynamically generated clusters

Highlights

• New approach to the organization of dispersed decision system structure was proposed.

• The local knowledge bases, which are in a friendship relation, are grouped.

• The groups of local knowledge bases will be generated in a dynamic way.

• Process of elimination inconsistencies in knowledge are implemented in groups.

• Global decisions are made by using one of the methods for analysis of conflicts.

Abstract

This paper discusses the issues 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). In the paper a decision-making system is described. In this system, the classification process of the test object can be divided into several steps. In the first step, we investigate how particular classifiers classify a test object. We describe this using probability vectors over decision classes. We cluster classifiers with respect to similarities of the probability vectors. 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 clusters.

The paper proposes a new approach to the organization of the structure of a decision-making system, which operates on the basis of dispersed knowledge. In the presented system, the classifiers are combined into groups called clusters in a dynamic way. We seek to designate groups of classifiers that classify the test object in a similar manner. The groups of classifiers are not disjoint sets. We use overlapping clusters because this is a more suitable representation of classification compatibility. It is assumed that, if the classifier classifies the test object in an ambiguous way, it should belong to several clusters. Then, a process of the elimination of inconsistencies in the knowledge is implemented in the created groups. Global decisions are made by using one of the methods for the analysis of conflicts.

Introduction

The problem of making decisions on the basis of dispersed knowledge stored in many local knowledge bases is examined in this paper. This problem concerns complex knowledge bases in which the possibility of the cooperation of local bases in order to reach a common decision (a global decision) is very important. In this paper a new approach to the organization of a system structure that uses dispersed knowledge is proposed.

In the proposed approach, we assume that each local knowledge base is managed by one agent. This situation can be seen as a set of classifiers, where each of the classifiers has access to a different knowledge set. In order to take a global decision, agents are combined into groups. In the new approach described in this paper, groups are created in a dynamic way. In earlier papers [22], [23], [24], [30], [31], [32], [33], a multi-agent system was also considered, but that system has a different structure. In the approaches used previously, the static structure of the system or a dynamic structure of the multi-agent system in which groups of agents are disjoint were considered. In the system proposed in this paper, a dynamic structure in which groups of agents are not disjoint sets is considered. In the proposed system, we aim to designate homogeneous groups of agents. The agents which agree on the classification for a test object into the decision classes are combined in a group. Many real life applications are characterized by situations in which overlapping clusters would be a more suitable representation. Very often agents that take part in the negotiations are unable to take one, explicit decision. Sometimes a few decisions are acceptable to one decision unit. In this situation agents should belong to several clusters. From a more technical point of view, in order to define clusters for each agent, a probability vector over decision classes is determined. This vector describes the classification of a test object that is made by the agent. Then agents are clustered with respect to the similarities of probability vectors. In order to identify groups of agents, the concepts of a friendship relation and a conflict relation, which were introduced by Pawlak in the papers [16], [17], [18], are used.

In the paper, a multi-agent system with a hierarchical structure is used. Local decisions are taken based on the knowledge of agents from one group. For every cluster, a kind of combined information is determined. Since the sets of attributes, conditions on the basis of which agents classify the test object do not have to be disjoint, an inconsistency in knowledge can occur. Therefore, a method for the elimination of inconsistencies in the knowledge is discussed here. Finally, the test object is classified by voting among clusters, using the combined information from each of the clusters. The problem of conflict analysis arises because the inference is being conducted in groups of knowledge bases. By a conflict, we mean a situation in which conflicting decisions are taken for the specified set of conditions on the basis of the knowledge that is stored in different groups of knowledge bases. This paper discusses two methods of conflict analysis (proposed in earlier papers [31], [32], [33]) that allow inference in spite of the presence of conflicts.

The paper is organized as follows. The second section presents the related papers. The third section describes the structure of a decision-making system. This section is divided into three parts. The first part of this section contains a high-level design description of the systems structure and the second part contains more technical issues. The last part of the third section presents an example of creating clusters. The fourth section describes the method of the elimination of inconsistencies in the knowledge. The fifth section describes the method of conflict analysis. The sixth section shows a description and the results of experiments carried out using some data sets from the UCI repository. The article concludes with a short summary in the seventh section.