Fuzzy expert systems, in

In this Chapter, the steps necessary to develop a fuzzy expert system (FES) from the initial model design through to final system evaluation will be pre-sented. The current state-of-the-art of fuzzy modelling can be summed up informally as ‘anything goes’. What this actually means is that the developer of the fuzzy model is faced with many steps in the process each with many options from which selections must be made. In general, there is no specific or prescriptive method that can be used to make these choices, there are simply heuristics (‘rules-of-thumb’) which may be employed to help guide the process. Each of the steps will be described in detail, a summary of the main options available will be provided and the available heuristics to guide selection will be reviewed. The steps will be illustrated by describing two cases studies: one will be a mock example of a fuzzy expert system for financial forecasting and the other will be a real example of a fuzzy expert system for a medical application. The expert system framework considered here is restricted to rule-based systems. While there are other frameworks that have been proposed for processing information utilising fuzzy methodologies, these are generally less popular in the context of fuzzy expert systems. conception of the process being studied and hence fuzzy model is the notional representation of the process in terms of fuzzy variables, rules and methods that together define the input-output mapping relationship. In contrast, the term system (as in fuzzy expert system) is used to refer to the embodiment, re-alisation or implementation of the theoretical model in some software language or package. A single model may be realised in different forms, for example, via differing software languages or differing hardware platforms. Thus it should be realised that there is a subtle, but important, distinction between the eval-uation of a fuzzy model of expertise and the evaluation of (one or more of) its corresponding fuzzy expert systems. A model may be evaluated as accurately capturing or representing the domain problem under consideration, wherea