A dynamic method for discretization of continuous attributes

Abstract

A discretization technique converts continuous attribute values into discrete ones. Discretization is needed when classification algorithms require only discrete attributes. It is also useful to increase the speed and the accuracy of classification algorithms. This paper presents a dynamic discretization method, whose main characteristic is to detect interdependencies between all continuous attributes. Empirical evaluation on 12 datasets from the UCI repository shows that the proposed algorithm is a relatively effective method for discretization.