Study of Assigning Imperfect Attribute Values for Classifier

Abstract

One of the interesting and important fields of research in data mining is classification on imperfect data. Unknown value can appear in real-world data sets at the stage of data collection. These facts lead to the imperfection of the decision system. The attention of this paper is to study the methodologies in making a decision point on such incompletion of data sets. We found a well-known rough set (RS) based classification scheme: Learning from Examples based on Rough Sets (LERS) that could be treated missing data, numeric data, and inconsistent data set. In this study, we utilize two interpreted meaning of imperfect values: lost values and attribute-concept values. The classification system is illustrated using a case study of iris dataset from the UCI  repository. The system is intended to present a comprehensive view of assigning on imperfect attributes value that generates better result among lost and attribute-concept values.