Defect prediction aims at identifying software artifacts that are likely to exhibit a defect. The main purpose of defect prediction is to reduce the cost of testing and code review, by letting developers focus on specific artifacts. Several researchers have worked on improving the accuracy of defect estimation models using techniques such as tuning, re-balancing, or feature selection. Ultimately, the reliability of a prediction model depends on the quality of the dataset. Therefore effort has been spent in identifying sources of noise in the datasets, and how to deal with them, including defect misclassification and defect origin. A key component of defect prediction approaches is the attribution of a defect to a projects release. Although developers might be able to attribute a defect to a specific release, in most cases a defect is attributed to the release after which the defect has been discovered. However, in many circumstances, it can happen that a defect is only discovered several releases after its introduction. This might introduce a bias in the dataset, i.e., treating the intermediate releases as defect-free and the latter as defect-prone. We call this phenomenon a “sleeping defect”. We call “snoring” the phenomenon in which classes are affected by sleeping defects only, that would be treated as defect-free until the defect is discovered. In this work, we analyze, on data from more than 4,000 bugs and 600 releases of 20 open source projects from the Apache ecosystem for investigating: 1)the magnitude of the sleeping defects, 2) the magnitude of the snoring classes, 3)if snoring impacts the evaluation of classifiers, 4)if snoring impacts classifier accuracy, and 5)if removing the last releases of data is beneficial in reducing the negative impact of the snoring noise on classifiers accuracy. Our results show that, on average across projects: 1)most of the defects in a project slept for more than 19% of the existing releases, 2)the missing rate is more than 50% unless we remove more than 20% of the releases, 3) the relative error in measuring the classifier accuracy achieved by using a dataset with snoring is about 100% in all accuracy metrics other than AUC, 4) the presence of snoring decreases the accuracy in each of the 15 classifiers, in each of the 6 accuracy metrics. For instance, Recall, F1, Kappa and Matthews decreases by about 80%, and 5) removing one release of data is better than removing no data in all accuracy metrics. For instance, Recall, F1, Kappa and Matthews increase by about 30%.
| Identifer | oai:union.ndltd.org:CALPOLY/oai:digitalcommons.calpoly.edu:theses-3464 |
| Date | 01 June 2019 |
| Creators | Ahluwalia, Aalok |
| Publisher | DigitalCommons@CalPoly |
| Source Sets | California Polytechnic State University |
| Detected Language | English |
| Type | text |
| Format | application/pdf |
| Source | Master's Theses |
Page generated in 0.0018 seconds