Classifier PGN: Classification with High Confidence Rules

Abstract

Associative classifiers use a set of class association rules, generated from a given training set, to classify new instances. Typically, these techniques set a minimal support to make a first selection of appropriate rules and discriminate subsequently between high and low quality rules by means of a quality measure such as confidence. As a result, the final set of class association rules have a support equal or greater than a predefined threshold, but many of them have confidence levels below 100%. PGN is a novel associative classifier which turns the traditional approach around and uses a confidence level of 100% as a first selection criterion, prior to maximizing the support. This article introduces PGN and evaluates the strength and limitations of PGN empirically. The results are promising and show that PGN is competitive with other well-known classifiers.

ACM Computing Classification System (1998): H.2.8, H.3.3.