Academic Commons

Reports

Testing and Validating Machine Learning Classifiers by Metamorphic Testing

Xie, Xiaoyuan; Ho, Joshua W. K.; Murphy, Christian; Kaiser, Gail E.; Xu, Baowen; Chen, Tsong Yueh

Machine Learning algorithms have provided important core functionality to support solutions in many scientific computing applications - such as computational biology, computational linguistics, and others. However, it is difficult to test such applications because often there is no "test oracle" to indicate what the correct output should be for arbitrary input. To help address the quality of scientific computing software, in this paper we present a technique for testing the implementations of machine learning classification algorithms on which such scientific computing software depends. Our technique is based on an approach called "metamorphic testing", which has been shown to be effective in such cases. Also presented is a case study on a real-world machine learning application framework, and a discussion of how programmers implementing machine learning algorithms can avoid the common pitfalls discovered in our study. We also conduct mutation analysis and cross-validation, which reveal that our method has very high effectiveness in killing mutants, and that observing expected cross-validation result alone is not sufficient to test for the correctness of a supervised classification program. Metamorphic testing is strongly recommended as a complementary approach. Finally we discuss how our findings can be used in other areas of computational science and engineering.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-002-10
Published Here
June 7, 2011