Knowledge Discovery in Databases and Decision Support

Two types of decision support systems (DSS) are discussed in Chapter 2: model-oriented and data-oriented. The assumption underlying model-oriented support is that accurate models of the decision problem are available or can be constructed prior to decision making. This approach includes optimization, simulation, decision analytic and statistical models. Various statistical methodologies can be used both to construct models and to estimate their parameters. Traditionally, these methods have been ap-plied to relatively small data sets; the resulting models were used to verify given hy-potheses, identify relationships among variables, and formulate predictions and sce-narios. Statistical methods require significant expertise in their application and interpreta-tion of results. These methods are difficult to use when there is little explicit knowl-edge about the issues and when the problems are ill-defined but described by large or very large amount of data. Therefore, these methods are normally used by analysts and researchers rather then directly by decision makers. While one may suspect that valuable knowledge is buried in the data, it was not until recently, that tools for knowledge extraction and presentation in an easily readable form have became avail-able. Three types of developments made data-oriented DSSs viable and effective. Artifi-cial intelligence (AI) adopted many statistical methods and embedded them in systems that efficiently search for patterns, derive rules, classify data according to some higher level concepts, and construct models from very large databases (Briscoe and Caelli 1996; Kohavi 1998). AI researchers have also developed methods to deal with such