CONCLUSIONS

Chapter XIX - Data Mining in Designing an Agent-Based DSS
Data Mining: Opportunities and Challenges
by John Wang (ed)
Idea Group Publishing 2003

	Brought to you by Team-Fly

A dynamic DSS that works efficiently, i.e., that does not constantly interrupt users with information requirements they cannot satisfy, must be able to identify the relationship among the characteristics of consults and domains. By using data-mining techniques it was possible to define a discriminating function to classify the system domains into two groups: those that can probably provide an answer to the information requirement made to the system, and those that can not do that. From this discriminating function, the system knowledge base was designed, which stores the values of the parameters required by such a function.

On the other hand, the system needs to learn from the errors it could make during its operation so that it can decrease the number of consulted domains in each information requirement presented to the system. The use of data mining allowed the definition of a data structure that is convenient for analyzing the system operation results and, according to that, for designing a cases base to store the information associated to the quality of each performed search.

Moreover, the application of data mining to the cases base allowed the specification of rules to settle relationships among the stored cases with the aim of inferring possible causes of error in the domains classification. In this way, a learning mechanism was designed to update the knowledge base and thus improve the already made classification as regards the values assigned to the discriminating function.

The selection mechanism designed using data-mining techniques is feasible to be implemented by means of agents technology. For this purpose, the roles required to operate the system have been identified (Jennings, 2000), and each of them was assigned to be under the responsibility of different software agents. Mobile agents take consults to the domains identified by an intelligent agent called a router agent. This agent is responsible for the classification and learning mechanism designed in this work.

In this way, it has been possible to design an agent-based architecture of a dynamic DSS that satisfies the main functionality specified for this system, i.e., to guide information requirements from users to the domains that offer the greatest possibility of answering them.

	Brought to you by Team-Fly