This chapter presents two approaches to data mining based on rough sets. In both cases generalizations of the original rough set theory are used. The first one, called VPRSM, may be used, for example, to acquire decision tables. The second approach, exemplified by LERS, is used for rule generation. Formed rules may be used for classification of new, unseen cases or for interpretation of regularities hidden in the input data.
When VPRSM is used for the generation of decision tables, global computations are involved, i.e., all attributes are taken into account in operations such as reducing the attribute set or analysis of significance of attributes. On the other hand, LERS usually uses a local approach. Computations in LERS involve attribute-value pairs rather than entire attributes, since the main task is to form rules containing attribute-value pairs.
Data mining based on rough set theory or, more exactly, on generalizations of rough set theory, were successfully used for more than a decade. A number of real-life projects, listed in this chapter, confirmed viability of data mining based on rough set theory.