Adaptive discretizer for machine learning based on granular computing and rough sets

Abstract

Machine-learning approaches based on granular computing and rough sets are good at dealing with discrete values and symbolic data. In this paper, a novel adaptive discretizer is proposed to discretize attributes with continuous values so that granular computing and rough set theory can avoid dealing with huge number of continuous values. It is demonstrated that this adaptive discretizer can improve quality of reducts and reduce the number of basic granules in an information system with continuous attributes. The experimental results on benchmark data sets show that the adaptive discretizer can improve the decision accuracy for the machine learning approaches based rough sets.