Increasing the economic value of expert systems: Theoretical and practical considerations

Vijay Sukumar Mookerjee, Purdue University

Abstract

This dissertation deals with methods to design expert systems with economic objectives. The problem addressed is that current methods to build expert systems fail to design systems that maximize system value, determined by the value of the solutions provided by the system minus the cost of generating these solutions. Expert system value is a critical consideration for these systems to be widely used in business. However, expert system technology has not explicitly focussed on system value. We address two distinct, but related problems. In the first problem, the objective is to minimize the costs of providing information needed by a system to reach solutions (recommendations). Many expert systems will fail to deliver the maximum possible value to their investors because little attention has been paid to the cost of providing these systems with the information they require to make a decision. Minimizing information costs without affecting the decisions made by the system can reduce the cost of operating the system and thereby increase system value. We develop an algorithm to determine a cost minimizing information acquisition strategy for an existing system and show how a specific strategy can be implemented. Because of the computational complexity of the algorithm, we present a simpler, heuristic solution to the problem. Experimental results indicate that the heuristic produces near-optimal strategies which are considerably better than strategies used by Prolog implementations of the same system. In the second problem, we develop an approach to build an economically optimal expert system. Recently, there has been a growing interest in the use of induction, a machine learning technique, to develop expert systems. Expert systems built using induction are termed as "Inductive Expert Systems." Induction offers an alternative to the costly process of manually extracting human knowledge to develop expert systems. The approaches used to build inductive systems seldom attempt to maximize system value. We present an induction algorithm that develops inductive expert systems with the objective of maximizing system value. We compare the performance of our algorithm to two existing induction algorithms in a simulation study. Our results indicate that system value can be significantly increased using our approach.

Degree

Ph.D.

Advisors

Santos, Purdue University.

Subject Area

Management

COinS