Cost minimization attribute reduction based on mutual information

The classical rough set attribute reduction is mainly based on maintaining three unvarying regions: positive region, boundary region, and negative region. In the decision rough set model, the reduction procedure for adding or deleting an attribute is no longer monotonous, and as such the three regions will not be kept unchanged simultaneously. In decision theoretic rough set model, cost should be taken more consideration when making decisions. Therefore, in this paper, an attribute reduction algorithm based on minimizing the cost is proposed. To quantitatively evaluate the classification power of selected attribute subset in the decision making, mutual information is introduced. To avoid the strong correlation between selected attributes, the significance of an attribute is computed from both maximum relevance and maximum significance, which can ensure the selected attributes to have strong classification ability but weak correlation. Experiments show the effectiveness of our proposed method.