Attribute reduction based on improved discernibility matrix

The attribute reduction based on information entropy is different to that based on positive region in inconsistent information system. The problem of discernibility matrix in algebra view is analyzed, and an new discernibility matrix based on information entropy is proposed in this paper. This algorithm considers whether the objects compared are consistent, analyses in detail the degree of inconsistency and the distributing proportion of their conditional equivalent classes in decision classes, and the reduction based on information entropy is acquired finally. The theoretic analysis and simulation instance shows that this algorithm is feasible and effective in practice.