Feature-oriented defect prediction

Software errors are a major nuisance in software development and can lead not only to reputation damages, but also to considerable financial losses for companies. Therefore, numerous techniques for predicting software defects, largely based on machine learning methods, have been developed over the past decades. These techniques usually rely on code and process metrics in order to predict defects at the granularity of typical software assets, such as subsystems, components, and files. In this paper, we present the first systematic investigation of feature-oriented defect prediction: the prediction of defects at the granularity of features---domain-oriented entities abstractly representing (and often cross-cutting) typical software assets. Feature-oriented prediction can be beneficial, since: (i) particular features might be more error-prone than others, (ii) characteristics of features known as defective might be useful to predict other error-prone features, (iii) feature-specific code might be especially prone to faults arising from feature interactions. We present a dataset derived from 12 software projects and introduce two metric sets for feature-oriented defect prediction. We evaluated seven machine learning classifiers with three different attribute sets each, using our two new metric sets as well as an existing metric set from the literature. We observe precision and recall values of around 85% and better robustness when more diverse metrics sets with richer feature information are used.