Boolean Separators and Approximate Boolean Classifiers 

Peter L. Hammer 
Irina I. Lozina

Abstract.   A simple technique is proposed for associating to a binary dataset a set of synthetic variables 
(called Boolean separators), some of which - if used either alone or in conjunction with the original variables - 
can enhance the accuracies of various frequently used machine learning / data mining methods. An iterative 
application of this technique is proposed for the generation of approximate Boolean classifiers, which are shown 
to increase the accuracy of each of the examined classification methods on each of the examined benchmark dataset.