Attribute extraction from mixed-mode data

In the rough set theory, discretization is a new feature extraction process, and reduction is an attribute selection process. The global discretization problem assumes that all conditional attributes of the decision table are continuous, while the reduct problem assumes that all conditional attributes are discrete. In this paper we firstly point out that under the context of mixed-mode data, these two problems are two extremes of a generalized problem, namely, the attribute extraction problem. Then we extend an existing approach to convert the attribute extraction problem of mixed-mode data into the traditional reduct problem, and prove that this conversion essentially incurs no loss of information. Moreover, this conversion is applicable for both decision tables and information tables. Examples are given to illustrate our scheme further.