A fuzzy rule-based decision support system environment

Abstract

This thesis proposes a fuzzy rule-based decision support system environment aimed at handling semi-structured decision problems at higher organisational levels. A three layered decision modelling formalism based on decision models, decision stages and decision rules is proposed. Decision models, at the top of the hierarchy, represent decomposed decision problems and specification of their composition using decision stages. Decision stages represent the smallest decision problems, and consist of a specification of probable alternatives and decision rules to be employed in constraining and ranking these probable alternatives. Lastly, decision rules, at the bottom of the hierarchy, represent the basic modelling knowledge as a high level declarative fuzzy rule. Imprecision and uncertainty in decision rules and decision parameters is represented using the concepts of fuzzy sets, possibility distributions, and necessity-possibility measures. The decision support environment presented in this thesis integrates a number of concepts from AI and ES into the generic DSS framework. The knowledge system comprises of modelling knowledge and environmental knowledge components. The modelling knowledge component, based on the formalism mentioned above, represents complete domain specific knowledge. The environmental knowledge component comprises of procedural and factual knowledge and it is represented by Prolog-like rules with extensions to represent imprecision and uncertainty using fuzzy sets, possibility distributions, and necessity- possibility measures. The language system provides a non-procedural command oriented language interface for model execution and information retrieval. The decision models can be activated for execution either by specifying the name of the decision model or by formulating new models. A fuzzy query language provides the facilities to retrieve information from the database and the knowledge base. The feasibility of the proposed environment is demonstrated by an implementation in Prolog on a VAX 11/750. The effectiveness of this environment in solving semi-structured decision problems is illustrated with an example.