Tolerance Rough Sets Using Grey Relational Analysis with Accumulated Generating Operation for Classification Problems

摘要

Tolerance rough sets (TRS) can operate effectively on continuous attributes and have been widely applied to pattern classification. This study uses grey relational analysis (GRA) to measure the similarity of any two patterns for TRS rather than the commonly-used distance function, since GRA can effectively measure relationships among data sequences. The accumulated generating operation (AGO) used to generate new features rather than the original features are incorporated into GRA. Since AGO can identib) potential regularity hidden in a data sequence, it is interesting to examine if such a combination can effectively improve classification performance compared to traditional TRS with a simple distance function. For this, a novel AGO with feature selection is further proposed. To yield high classification performance, a genetic-algorithm-based learning algorithm was designed to generate the AGO-based tolerance class of a pattern. Experimental results on several real-world data sets show that the proposed classification method performs well in comparison with other rough-set-based methods.