A Comparative Study on Strategies of Rule Induction for Incomplete Data Based on Rough Set Approach

Rough set based rule induction approaches have been studied intensively during past few years. However, classical rough set model cannot deal with incomplete data sets. There are two main categories dealing with this problem: the preprocessing methods and the extensions of rough set model. This paper focuses on the comparison of three strategies for dealing with incomplete data containing three preprocessing methods and one extended discernibility matrix method. These three methods only different when building the discernibility matrix, and they have the same rule induction method. The result shows that some preprocessing methods are stable and relatively effective, while the extended discernibility matrix method is not very effective in dealing with incomplete data.