Automatic Defect Categorization

Defects are prevalent in software systems. In order to understand defects better, industry practitioners often categorize bugs into various types. One common kind of categorization is the IBM's Orthogonal Defect Classification (ODC). ODC proposes various orthogonal classification of defects based on much information about the defects, such as the symptoms and semantics of the defects, the root cause analysis of the defects, and many more. With these category labels, developers can better perform post-mortem analysis to find out what the common characteristics of the defects that plague a particular software project are. Albeit the benefits of having these categories, for many software systems, these category labels are often missing. To address this problem, we propose a text mining solution that can categorize defects into various types by analyzing both texts from bug reports and code features from bug fixes. To this end, we have manually analyzed the data about 500 defects from three software systems, and classified them according to ODC. In addition, we propose a classification-based approach that can automatically classify defects into three super-categories that are comprised of ODC categories: control and data flow, structural, and non-functional. Our empirical evaluation shows that the automatic classification approach is able to label defects with an average accuracy of 77.8% by using the SVM multiclass classification algorithm.