Fuzzy Query by Sonalysts

White Paper: Knowledge Discovery and Data Mining

Introduction

Organizations throughout the world have begun to recognize that their vast stores of historical and active data represent a critical resource. These resources comprise the intellectual property of the organization, both its underlying proprietary knowledge as well as an explicit picture of its business process models. These active and archival databases contain implicit relationships between crucial elements and data within the organization.

The assemblage of techniques used to discover these deep and nearly invisible relationships is called data mining and data discovery. As a consequence of this technique's ability to detect hidden variables and hidden dependencies in often vast collection of data, organizations of all sizes and missions are attempting to find "nuggets of gold", that is, undiscovered relationships in the data, that will boost profitability, improve corporate productivity, and give the organization an edge in today's highly competitive environment.

More and more, organizations are turning to their own data as a way of answering questions such as "How can we improve business performance?", "Where are we vulnerable to competitive attacks?", "How can we identify new markets for existing products?", "Where should we position ourselves for the next century?"

The Data Mining Process

Through data mining, large collections of corporate historical and active data are cleaned, organized, statistically analyzed, and then explored in order to reveal any deep and potentially profitable relationships. A data mining process should be able to produce a working model of the underlying data relationships. More often than not, however, the current generation of tools simply generates a report identifying any discovered dependencies among data elements.

The Data Mining Process

Rule Induction and Dynamic Fuzzy Models

Most data mining tools and techniques (with the exception of neural networks and our approach to
rule induction) produce a static report as the outcome of their data mining process. The rule
induction process, on the other hand, evolves a dynamic system model of the underlying system
through a set of rules.

The Idea of a Model

This is the most powerful method of data mining and knowledge discovery. With a rule based
system the knowledge mining engineer can:

Examine the actual relationships
Run and simulate the behavior of the system
Ask the system to explain its reasoning, and

Modify the evolved system to include additional rules or change the nature of one or more
evolved rules. Since evolved dynamic models can be regenerated, they are ideally suited to adaptive feedback systems and processes that are sensitive to rapid changes in the external world.

A Dynamic Fuzzy System

For more information E-Mail: FuzzyQuery@Sonalysts.com

Fuzzy Systems Solutions
Sonalysts Inc.
215 Parkway North
Waterford, CT 06385
Tel: 800-526-8091 Fax: 860-447-8883

© 2003 Sonalysts Inc. All rights reserved.