Perspective of Rough Set Theory in Feature Selection and its Application

Abstract

Rough Set hypothesis was presented by Z. Pawlak in 1982 as a scientific instrument for information examination. From that point forward it has been utilized to handle unverifiable information in applications of Artificial Intelligence with numerous applications in the field of Knowledge Discovery in Databases (KDD) among them feature selection, discretization, feature reduction and clustering are common examples. A rough set theory utilization is a determination of features. Equivalence relations can be found among a few dataset instances, and some of them can be chosen to shape another subset to be utilized as a part of future examinations. In this way, instance (feature or attribute) reduction or selection includes separating rudimentary pieces from the dataset in view of an equivalence relation. As of late, Basu outlined a numerical model, named rough finite state automata, which perceives such rough sets and is believed to end up being of awesome significance to the researchers in the field of data analysis in near future. RFSA introduced by Basu is a tool which can perform analysis of uncertain data and recognize rough languages. It is a new concept, with not much research done, but we feel will prove to be useful in the long run. In this proposition, we have led research on the utilization of hypothesis of rough sets in a few Knowledge Discovery undertakings. Thus, various tools used in the field of extraction of precise data in the field of rough sets are analyzed and compared with each other. We then provide an application of Rough Finite State Machine in various real-world data sets, which contain imprecise and vague information, for a better classification. A feature selection method based on Rough Sets is searched while analyzing different techniques of dependency function based reduction algorithms. It is used as it provides a better and optimal classification of imprecise data as compared to rest of the techniques. The algorithm is then applied to various domains, and the results are thus compared.