Rough Set Based Classi cation Methods and Extended

In a practical sense, database mining applications involve semi-automatic data analysis methods that help users to discover some non-trivial knowledge. We are interested in four kinds of database mining queries: association, hypothesis testing, classiication, and characterization. Since these queries can be viewed as special cases of data dependencies in a relation, the rough set theory provides a sound basis for investigating database mining. We propose four classiication methods in an algebraic approximation space and compare them in terms of their classiication quality. We also show that these methods are in fact special cases of classiication methods in a probabilistic approximation space. The result of a classiication method, regardless of type of space on which it is based, is called a decision algorithm, which can be either consistent or inconsistent. In order to keep a decision algorithm from becoming obsolete, some kind of frequency information must be associated with it. This information is called incremental information. The incremental information is also used for interpreting an inconsistent decision algorithm deter-ministically or nondeterministically. We extend the notion of decision tables from the rough set theory to accommodate incremental information for the classiication of objects.