Defect, defect, defect: defect prediction 2.0

摘要: Defect prediction has been a very active research area in software engineering [6--8, 11, 13, 16, 19, 20].In 1971, Akiyama proposed one of the earliest defect models using Lines Code (LOC) [1]: "Defect = 4.86 + 0.018LOC."Since then, many effective new and metrics have proposed. For models, typical machine learners regression algorithms such as Naive Bayes, Decision Tree, Linear Regression are widely used. On other hand, Kim et al. cache-based model bug occurrence properties [9]. Hassan change entropy to effectively predict defects [6]. Recently, Bettenburg Multivariate Adaptive Splines improve by learning from local global together [4].Besides LOC, for Among them, source code history used yield reasonable accuracy. example, Basili [3] Chidamber Kemerer metrics, Ohlsson [14] McCabe's cyclomatic complexity prediction. Moser [12] number revisions, authors, past fixes, age file predictors. micro interaction (MIMs) [10] quality measures [15] proposed.However, there is much room 2.0. First all, understanding actual causes necessary. Without we may reach nonsensical conclusions results [18]. Many proposed, but successful application cases practice scarcely reported. To be more attractive developers practice, it desirable finer granularity levels line or even keyword level. Note that static finders FindBugs [2] can identify potential bugs level, find them useful practice. Dealing with noise data become an important issue. Bird identified non-neglectable [5]. This poor and/or meaningless results. Cross-prediction highly desirable: projects limited training data, necessary learn sufficient projects, apply those projects. However, Zimmermann [21] cross-project challenging problem. Turhan [17] analyzed Cross-Company (CC) Within-Company (WC) prediction, confirmed reuse CC directly companies' software.Overall, interesting promising area. still challenges problems addressed. Hopefully, this discussion calls solutions ideas address these challenges.