An empirical study of experience-based software defect content estimation methods

Capture-recapture models and curve-fitting models have been proposed to estimate the remaining number of defects after a review. This estimation gives valuable information to monitor and control software reliability. However, the different models provide different estimates making it difficult to know which estimate is the most accurate. One possible solution is to, as in this paper, focus on different opportunities to estimate intervals. The study is based on thirty capture-recapture data sets from software reviews. Twenty of the data sets are used to create different models to perform estimation. The models are then evaluated on the remaining ten data sets. The study shows that the use of historical data in model building is one way to overcome some of the problems experienced with both capture-recapture and curve-fitting models, to estimate the defect content after a review.