COMPOSITE BOOLEAN SEPARATORS FOR DATA ANALYSIS

Peter L. Hammer   and   Irina I. Lozina

Abstract. In  [12],  we  proposed  a  simple procedure for generating artificial Boolean variables. 
In  this paper,  we  present  a  formal  description  of  the  procedure and apply the new variables 
to different problems in machine-learning / data-mining. In particular, we demonstrate the usefulness 
of these concepts by showing how the introduction of artificial variables can enhance the accuracy of 
classification systems; we employ the new variables for identifying misclassified  observations  and  
examine  how  deletion  of  such  observations  and  reversal  of their class  influence  the  classification  
accuracy;  we  apply  the  new  artificial  variables  to  the attribute selection  problem,  i.e.,  to  
the  problem  of  identifying  informative  subsets  of  the  original attributes. All the results have been 
tested on eight publicly available datasets and validated by five well-known machine-learning / data-mining methods.