A Comparative Analysis of Discretization Techniques in Machine Learning
Raed Alazaidah
Abstract
Data discretization is a very significant pre-process step that has a great impact on the predictive performance of several machine learning algorithms. This step aims to convert continuous attribute with large number of real values into a finite set of intervals with minimum data loss. Several techniques have been proposed to accomplish this task, without a proof which one is better with respect to data characteristics. Hence, this paper investigates and evaluates fifteen discretization techniques using two popular Associative Classification (AC) based classifiers (CBA and CBA2), with respect to Accuracy metric. The results revealed that both Ameva and Chi-2 techniques show a superior performance using five different datasets comparing with the other considered techniques.