Knowledge quality functions for rule discovery

Abstract

The Department of Defense (DoD) possesses tremendous amounts of data stored in many large databases. Due to the size of these databases, humans are incapable of efficiently discovering interesting and useful patterns so an automated data-mining tool is necessary. Output in the form of production rules, ie., "Ify Then x," is preferred because they are understandable by humans and support decision making processes. This thesis investigates the manner in which data-mining systems discover useful, interesting, but currently unavailable knowledge. The search and evaluation process, guided by a knowledge quality function, is the key task of a data-mining system This thesis evaluates three knowledge quality functions taken from the literature. Each knowledge quality function discovers new and interesting sets of rules reflecting different characteristics of knowledge. DoD applications are suggested for each of the knowledge quality functions.