Metric based attribute reduction in decision tables

In an information system, each subset of attributes determines knowledge structure on the set of objects, in which each element is an equivalence class. Thus, a metric which is defined on knowledge structures is established on the attribute sets. Once a metric is established, we can use the metric to measure attributes distance, cluster and discover important attributes. As a result, effective algorithms are constructed to solve attribute reduction in information systems. With metric on knowledge structures based on the Jaccard distance between two finite sets, this paper proposes a new method for attribute reduction in decision table. The paper proves theoretically and experimentally that this metric method is more effective than other methods based on conditional Shannon entropy.