Use of data analytics to build intuitive decision models – an approach to indirect derivation of criteria weights based on discordance related preferential information

Abstract

ABSTRACT Data on past and current decisions can be utilised to enhance the decision-making process by automating decisions or making problem solving more intuitive. Data is either extracted from distributed sources and repositories, or obtained with the regression analysis from holistically assessed alternatives and human judgements. One of possible advanced approaches to encourage intuitive decision-making aims at inferring criteria weights of the decision model with regard to correlations between preferential parameters, in such a way that objective inner information on alternatives is consolidated with personal knowledge and experience. This is relevant because the task of specifying criteria weights is cognitively demanding and represents a key aspect of each decision model. The paper first discusses the notation and infrastructure to exchange decision models and handle preferential information underlying the mechanisms of indirect weight derivation. As the main contribution of the research, a method for the inference of criteria weights from veto-related information is proposed, with which selective strengths of veto degrees are calculated to compare the magnitudes of veto, while strengths of veto assessments are used to determine the influence of veto on the deterioration of rankings or categories into which alternatives are sorted, respectively. Strengths of non-compensatory veto criteria are then projected into compensatory weights. The experimental study reveals the characteristics of indirectly derived criteria weights and the influence of veto. Several quality factors are considered, such as the validity of weights, accuracy of results, richness of information and ability to discriminate conflicting alternatives. Weights are also compared to standard ROC and RS surrogate weights. The approach is generalised to both common decision-making problematics of ranking and sorting.