Towards implementing defect prediction in the software development process

Defect prediction is a method of identifying possible locations of software defects without testing. Software tests can be laborious and costly thus one may expect defect prediction to be a first class citizen in software engineering. Nonetheless, the industry apparently does not see it that way as the level of practical usages is limited. The study describes the possible reasons of the low adoption and suggests a number of improvements for defect prediction, including a confusion matrix-based model for assessing the costs and gains. The improvements are designed to increase the level of practitioners acceptance of defect prediction by removing the recognized by authors implementation obstacles. The obtained predictors showed acceptable performance. The results were processed through the suggested model for assessing the costs and gains and showed the potential of significant benefits, i.e. up to 90% of the overall cost of the considered test activities.