ESRS: a case selection algorithm using extended similarity-based rough sets

Abstract

A case selection algorithm selects representative cases from a large data set for future case-based reasoning tasks. This paper proposes the ESRS algorithm, based on extended similarity-based rough set theory, which selects a reasonable number of the representative cases while maintaining satisfactory classification accuracy. It also can handle noise and inconsistent data. Experimental results on synthetic and real sets of cases showed that its predictive accuracy is similar to that of well-known machine learning systems on standard data sets, while it has the advantage of being applicable to any data set where a similarity function can be defined.