5 INTELLIGENT MODELING AND MODEL MANAGEMENT and to their management makes lots of sense because some of the tasks involved (e.g.,

~~- 15.5 INTELLIGENT MODELING AND MODEL MANAGEMENT and to their management makes lots of sense because some of the tasks involved (e.g.,

Adding intelligence to the process of modeling (building models or using existing models)

modeling and selecting models) require considerable expertise. The topics of intelligent modeling and intelligent model management attracted significant academic attention during the 1990s (Blanning, 1993; Chang et aI., 1993) because the potential benefits could be substantial. However, it seems that the implementation of such integration is fairly difficult and slow. For a survey of the early approaches, see Suh et al. (1995) and Eom (1999). However, the introduction of Web Services may solve many of the integration problems (see Cerami, 2002). For a detailed study on computer-based modeling environments from the perspectives of modelers (analysts) and model users (decision-makers), see Wright et al. (1998).

ISSUES IN MODEL MANAGEMENT

Wu (2000) developed a model management system for helping nonexperts make decisions related to test construction. The system consists of four components: problem analysis, model-type selection, model formulation, and model solver. This system proved to be very user-friendly and efficient.

We discuss here four similar interrelated subtopics of model management: prob- lem diagnosis and selection of models, construction of models (formulation), use of models (analysis), and interpretation of the output of models.

PROBLEM DIAGNOSIS AND SELECTION OF MODELS

Several commercial ES are now helping to select appropriate statistical models (e.g., Statistical Navigator, at static.elibrary.com). Goul et al. (1984) have developed a model selection of ES to be used in mathematical programming, and Courtney et al. (1987) have produced an expert system for problem diagnosis. Liang and Konsynski (1993) have suggested using analogy as a source of knowledge for modeling. Dutta (1996) claims that model selection is a major area of AI and optimization integration. Venkatachalam and Sohl (1999) have presented an application of ANN for forecasting model selection. Lu et al. (2000) have proposed a guidance framework for designing intelligent systems to help a typical decision-maker in selecting the most appropriate method for solving various multiobjective decision-making problems.

CONSTRUCTION OF MODELS

The construction of models for decision-making involves the simplification of a real- world situation so that a less complex representation of reality can be made. Models can be normative or descriptive, and they can be used in various types of computerbased information systems (especially DSS). Finding an appropriate balance between simplification and representation in modeling requires expertise. The definition of the problem to be modeled, the attempt to select a standard model (e.g., linear programming), the data collection, the model validation, and the estimation of certain parameters and relationships are not simple tasks. For instance, data can be tested for suitability for a certain statistical distribution (e.g., does the arrival rate in queuing follow a Poisson distribution?).

The ES could guide the user in selecting an appropriate test and interpreting its results,

which in turn can help in appropriate modeling of the situation.

Knowledge-discovery techniques, such as decision-rule discovery, offer intelligent

PART V IMPLEMENTING MSS IN THE E-BuSINESS ERA

Such an approach can minimize the effort required for model builders (or analysts) to model the decision-making processes of decision-makers. For details see Bolloju (1999).

USE OF MODELS

Once models are constructed, they can be put to use. The application of models may require some judgmental values (e.g., setting an alpha value in exponential smoothing). Experience is also needed to conduct a sensitivity analysis as well as to determine whet constitutes a significant difference (e.g., is project A really superior to project B ?). Expert systems can provide the user with the necessary guidelines for the use of models.

INTERPRETATION OF RESULTS

Expert systems are able to provide explanations of the models used and interpretations of the derived results. For example, an ES can trace anomalies in. the data. Furthermore, sensitivity analysis may be needed, or it may be necessary to convert information to a certain format. An ES can advise in all of the above cases,