DEVELOPMENT OF TECHNIQUE FOR STRUCTURING OF GROUP EXPERT ASSESSMENTS UNDER UNCERTAINTY AND INCONCISTANCY

Abstract

Context. The issues of structuring group expert assessments are considered in order to determine a generalized assessment under inconsistency between expert assessments. The object of the study is the process of synthesis of mathematical models of structuring (clustering, partitioning) of expert assessments that are formed within the framework of Shafer model under uncertainty, inconsistency (conflict). Objective. The purpose of the article is to develop an approach based on the metrics of theory of evidence, which allows to identify a number of homogeneous subgroups from the initial heterogeneous set of expert judgments formed within the framework of the Shafer model, or to identify experts whose judgments differ significantly from the judgments of the rest of the group. Method. The research methodology is based on the mathematical apparatus of theory of evidence and cluster analysis. The proposed approach uses the principles of hierarchical clustering to form a partition of a heterogeneous (inconsistent) set of expert evidence into a number of subgroups (clusters), within which expert assessments are close to each other. Metrics of the theory of evidence are considered as a criterion for determining the similarity and dissimilarity of clusters. Experts’ evidence are considered consistent in the formed cluster if the average or maximum (depending on certain initial conditions) level of conflict between them does not exceed a given threshold level. Results. The proposed approach for structuring expert information makes it possible to assess the degree of consistency of expert assessments within an expert group based on an analysis of the distance between expert evidence bodies. In case of a lack of consistency within the expert group, it is proposed to select from a heterogeneous set of assessments subgroups of experts whose assessments are close to each other for further aggregation in order to obtain a generalized assessment. Conclusions. Models and methods for analyzing and structuring group expert assessments formed within the notation of the theory of evidence under uncertainty, inconsistency, and conflict were further developed. An approach to clustering group expert assessments formed under uncertainty and inconsistency (conflict) within the framework of the Shafer model is proposed in order to identify subgroups within which expert assessments are considered consistent. In contrast to existing clustering methods, the proposed approach allows processing expert evidence of a various structure and taking into account possible ways of their interaction (combination, intersection).