An new algorithm-based rough set for selecting clustering attribute in categorical data

Abstract

Several algorithms strategies based on Rough Set Theory (RST) have been used for the selection of attributes and grouping objects that show similar features. On the other hand, most of these clustering techniques cannot deal tackle partitioning. In addition, these processes are computationally complexity and low purity. In this study, the researcher looked at the limitations of the two rough set based techniques used, Information-Theoretic Dependency Roughness (ITDR) and Maximum Indiscernible Attribute (MIA). They also proposed a novel method for selecting clustering attributes, Maximum mean Attribute (MMA). They compared the performance of MMA, ITDR and MIA technique, using UCI and benchmark datasets. Their results validated the performance of the MMA with regards to its purity and computational complexity.