Accumulated Cost Based Test-Cost-Sensitive Attribute Reduction

As a generalization of the classical reduct problem, test-costsensitive attribute reduction aims at finding a minimal test-cost reduct. The performance of an existing algorithm is not satisfactory, partly because that the test-cost of an attribute is not appropriate to adjust the attribute significance. In this paper, we propose to use the test-cost sum of selected attributes instead and obtain a new attribute significance function, with which a new algorithm is designed. Experimental results on the Zoo dataset with various test-cost settings show performance improvement of the new algorithm over the existing one.