Machine learning method for software quality model building
Maurício Amaral de Almeida and Stan Matwin
School of Information Technology and Engineering
University of Ottawa 150 Louis Pasteur, Ottawa Ontario, K1N 6N5 Canada
{malmeida,stan}@csi.uottawa.ca
Abstract
Software quality prediction can be cast as a concept learning problem. In this paper, we discuss the full cycle of an application of Machine Learning to software quality prediction. As it often happens in real-life applications, significant part of the project was devoted to activities outside the learning process: data acquisition, feature engineering, labeling of the examples, etc. We believe that in projects that reach out to real data (rather than rely on the prepared data sets from the existing repositories), these activities often decide about the success or a failure of the project. The method proposed here is applied to a set of real-life COBOL programs and some discussion on the results is presented.