Dynamic reducts as a tool for extracting laws from decisions tables

Information Granulation and Pattern Recognition

“…Hence, several techniques have been developed to select relevant reducts or their approximations. Among them is one based on so-called dynamic reducts [2,4]. The attributes are considered relevant if they belong to dynamic reducts with a sufficiently high stability coefficient, i.e., they appear with sufficiently high frequency in random samples extracted from a given information system.…”

“…Using rough sets [2,4,28,52] for feature selection was proposed in several contributions (see, e.g., [53,54]). The simplest approach is based on calculation of a core for a discrete attribute data set containing strongly relevant features and reducts containing a core plus additional weakly relevant features, such that each reduct is satisfactory for description of concepts in the data set.…”

“…Based on a set of reducts for a data set, some criteria for feature selection can be formed, for example, selecting features from a minimal reduct, i.e., a reduct containing a minimal set of attributes. Dynamic reducts were proposed to find a robust (well-generalizing) feature subset [2,4]. The selection of a dynamic reduct is based on the cross-validation method.…”

“…For real-life situations and limited-size data sets, other reducts (generally other feature subsets) might be better for generalizing prediction. A selection of a robust (generalizing) reduct, as a best open-loop feature subset, can be supported by introducing the idea of a dynamic reduct [2,4] or by an ensemble of classifiers defined by reducts [57].…”

“…We emphasize the special role of reducts in feature selection, including dynamic reducts [2,4,5]. Then, we present a short overview of a feature selection problem including openloop and closed-loop feature selection methods [9].…”