数据挖掘技术在计算机系统中的应用越来越广泛,对其软件做质量保证的测试势必日益受到重视。然而,传统的测试方法难以适应数据挖掘领域,因为其应用程序存在"oracle"问题。基于蜕变测试技术,提出数据挖掘领域的软件评测方法,并以一个聚类算法为具体案例展开研究分析,构造了一系列蜕变关系。实验表明,此方法不仅能达到核查程序正确性的目的,还可揭示算法的适用性。并且,此方法具有推广到其他领域的可行性。
As data mining more and more popular applied in computer system,the quality assurance test of its soft-ware would be get more and more attention.However,because of the existence of the "oracle" issue,the traditional test method is not ease fit for the application program in the field of the data mining.In this paper we present an approach called "metamorphic testing" for testing implementations of data mining algorithms on which such software depends.We present our findings in a case study of cluster mining algorithm,and identify a set of metamorphic relations for cluster.The approach has been shown to be effective for the algorithm,which not only serves the purpose of verification,but also reveals adaptability of algorithm.It extends beyond scientific computing and data mining into other domains.