Data-based revision of probability distributions in qualitative multi-attribute decision models

Article type: Research Article

Authors: Žnidaršič, Martin^{; *} | Bohanec, Marko

Affiliations: Department of Knowledge Technologies, Jožef Stefan Institute, Jamova 39, 1000 Ljubljana, Slovenija

Correspondence: [*] Corresponding author: Martin Žnidaršič. Tel.: +386 1 477 3366; E-mail: [email protected].

Abstract: An automatic data-based revision method of probabilistic multi-attribute decision models is proposed. Data-based revision of decision models is defined as follows: given an existing model and a set of data items, revise the model to match the data items. We propose and experimentally evaluate a method for the revision of probability distributions in qualitative hierarchical multi-attribute models of DEX methodology. The revision method is automatic, but limited to the modification of probability distributions in utility functions. The method is experimentally evaluated in an artificial domain. In all experiments, the classification accuracy of the revised model was improved and the changes of the model correctly reflected the simulated changes in the decision environment.

Keywords: data-based revision, decision modeling, multi-attribute decision making, probabilistic data-based modeling, knowledge refinement, machine learning

DOI: 10.3233/IDA-2005-9203

Journal: Intelligent Data Analysis, vol. 9, no. 2, pp. 159-174, 2005