Rule discovery from databases with decision matrices

Abstract

In this paper, we propose a new method which can compute all maximal general rules in relational databases. The method is based on the idea of a decision matrix and Boolean decision function. The problem of finding the maximal general rules is reduced to the problem of simplifying a group of associated boolean expression. The novel feature of our algorithm is its generation and use of explicit redundant knowledge rules instead of a set of minimum cover rules. Our algorithm generates all potential interesting and useful rules even though there is some overlapping between the examples covered by the rules. The test on some data sets demonstrates the effectiveness and high classification accuracy of a decision system based on redundant rules.