Intelligent Multiple Attribute Decision Making System

Liu Shiwei

Abstract

In this project, all the published Multiple Attribute Decision Making (MADM) methods by Hwang & Yoon (1981) have been studied, and the feasibility of developing an intelligent MADM system has been researched. And then, a fully functional prototype system has been developed. The prototype system was developed using Visual Basic programming language under Visual Studio platform. The executable software is able to run in any Microsoft System above Win95.

Because of the wide scope, many mathematical methodologies have been developed over the years to solve MADM problems. Generally, for a particular MADM problem, some MADM methods are more suitable and accurate than others. However, the method choosing expertise is not always available to the decision makers. Hence, the system should preferably have an intelligent component to guide the user to choose the most proper MADM methods. In addition, the system should have sufficient MADM methods implemented for decision maker's use.

In fulfillment of the two basic requirements, the developed system has two characteristics:

1. Flexibility to introduce or revise methods in the MADM methods pool.
2. Capability to guide decision maker in the method-choosing process.

In the current system, 10 of the 17 well established methods published by Hwang & Yoon (1981) had been implemented. As different methods require different types of problem information, these data input routines were designed as independent modules. This design facilitates the flexibility to introduce or modify MADM methods.

For the second characteristic, a Knowledge Based System (KBS) has been introduced to the system to provide guidance in the method selection process. This system does not purposely categorize decision makers into expert, intermediate and novice classes as previous designs did. A decision maker can simply invoke any MADM method at any time; the system will either solve or do not solve the problem based on problem information status (It is determined by the Bayesian network model). In this system, it is not necessary to classify decision makers based on familiarity to MADM.

In conclusion, this project demonstrated how the MADM system and KBS should be integrated together. It also shows the pleasant and smooth decision making process.