Browse Theses

This study investigates the feasibility of using relational database (RDB) technology to develop a knowledge based system (KBS) for model building. Using RDB technology to build a KBS has many advantages, including: (1) model building knowledge will be available to all modelers, regardless of the statistical or operations research software that is being used to estimate or optimize the model; (2) model building knowledge will be more consistent and more easily updated; (3) model building knowledge will be shareable across problem area domains; and (4) the KBS will be compatible with distributed processing environments.

The research methodology follows the engineering design paradigm suggested by Vogel and Wetherbe (1984). First, the design requirements for a KBS to support model building are stated. Both knowledge management requirements and model building requirements are considered. The focus is on how the KBS transfers knowledge to the user during the model building process. A model that describes this knowledge transfer is developed. The transfer is accomplished by processing a collection of condition/action statements.

The second step is to determine the specific functionalities that a relational database management system (RDBMS) can provide to satisfy the stated design requirements. Several methods are reviewed for using a RDBMS to process a collection of condition/action statements. A design is proposed that matches the requirements of the KBS and the capabilities of the RDBMS. This design is used to construct a KBS prototype. The prototype realizes all of the advantages cited above for using RDB technology, and hence demonstrates that it is both feasible and desirable to use RDB technology to develop a KBS for model building.

The infrastructure developed in this study is an important first step toward developing an integrated modeling environment. The proposed infrastructure must, however, be tested by both KBS users and problem domain experts to access its ultimate contribution. Test on users must be conducted to assess whether the model building process actually is improved (better models are developed or the models are developed more quickly) as a result of using such KBSs. Test on experts must be conducted to assess the ease with which knowledge can be both stored and maintained on such KBSs.