Formation of a Compact Reduct Set Based on Discernibility Relation and Attribute Dependency of Rough Set Theory

Large amount of data have been collected routinely in the course of day-to-day work in different fields. Typically, the datasets constantly grow accumulating a large number of features, which are not equally important in decision-making. Moreover, the information often lacks completeness and has relatively low information density. Dimensionality reduction is a fundamental area of research in data mining domain. Rough Set Theory (RST), based on a mathematical concept, has become very popular in dimensionality reduction of large datasets. The method is used to determine a subset of attributes called reduct which can predict the decision concepts. In the paper, the concepts of discernibility relation and attribute dependency are integrated for the formation of a compact reduct set which not only reduces the complexity but also helps to achieve higher accuracy of the system. Performance of the proposed method has been evaluated by comparing classification accuracy with some existing dimension reduction algorithms, demonstrating superior result.