Discretization method of continuous attributes based on decision attributes

Abstract

The attributes in rough set must be discretized, but the general theory on discretization did not think about the decision attribute adequately during discretization of data, as a result, it leads to several redundant rules and lower calculation efficiency. The discretization method of continuous attributes based on decision attributes which is discussed in this paper gives more attention to both significance of attributes and the decision attributes. The continuous attributes are discretized in sequence according to their significance. The result shows less breakpoints and higher recognition accuracy. The experiment on database Iris for UCI robot learning validates the feasibility of our method. Comparing the result with documents [6] and [11], the method given in this paper shows higher recognition accuracy and much less breakpoints. © 2010 Springer-Verlag.