A Rough Sets/Neural Networks Approach to Knowledge Discovery for the Development of Decision Support Systems

Abstract

The problem domain of decision support systems (DSS) has been characterised as ill-specified, influenced by many different sources, and subject to change. The complex and changing environment in which decision support systems are developed makes specification of the final system requirements difficult. Decision support systems need to be designed using an adaptive process of learning and experimentation. The emphasis in DSS development is on supporting the decision making process by improving the decision maker’ s understanding of the decision domain. This paper proposes a conceptual and a technological framework for adaptive development of decision support systems (DSS) through knowledge discovery from historical data. In this approach the users and developers gradually learn about their decision domain by an iterative process of extracting patterns from the database and building and refining predictive models using these patterns. A hybrid rough sets/neural networks framework that can facilitate this development is proposed. Finally a case study where the approach was applied in the development of a real-world decision support system is described and discussed. The framework can be applied to a variety of industrial and business decision problems where historical data exists.Keywordsdecision support systemsknowledge discoverysoft computingneural networksrough sets