Probabilistic Similarity-Based Reduct

The attribute selection problem with respect to decision tables can be efficiently solved with the use of rough set theory. However, a known issue in standard rough set methodology is its inability to deal with probabilistic and similarity information about objects. This paper presents a novel type of reduct that takes into account this information. We argue that the approximate preservation of probability distributions and similarity of objects within reduced decision table helps to preserve the quality of its classification capability.