Predicting software defect at the class level is important in defect prediction studies. In addition, class level defect prediction enables software team to have an idea on the possible number of defects in an upcoming product release prior to testing. Considerable effort is needed to investigate factors which influence the number of defects at the class level of a software product. In this paper, we investigate few factors, namely, time and defect velocity in relation to defect density of a class using a defect density correlation technique. An experiment was conducted on 69 open source Java projects containing 131,034 classes. Results show that the defect density achieved a correlation coefficient of 61.41%, defect introduction time achieved a correlation coefficient of -11.413%, and the defect velocity achieved 93.56%, respectively. Defect velocity is positively correlated with the number of defects at the class level.

View PDF No PDF? Try Arxiv!