Decision Queue Classifier for Supervised Learning Using Rotated Hyperboxes

Abstract

This article describes a new system for learning rules using rotated hyperboxes as individuals of a genetic algorithm (GA). Our method attempts to find out hyperboxes at any orientation by combining deterministic hill-climbing with GA. Standard techniques, such as C4.5, use hyperboxes that are aligned with the coordinate axes. The system uses the decision queue (DQ) as method of representing the rule set. It means that the obtained rules must be applied in specific order, that is, an example will be classify by the i-rule only if it doesn’t satisfy the condition part of the i-1 previous rules. With this policy, the number of rules is less because the rules could be one inside of another one. We have tested our system on real data from UCI repository. Moreover, we have designed some two-dimensional artificial databases to show graphically the experiments. The results are summarized in the last section.