Nowadays, software systems are essential for businesses, users and society. At the same time such systems are growing both in complexity and size. In this context, developing high-quality software is a challenging and expensive activity for the software industry. Since software organizations are always limited by their budget, personnel and time, it is not a trivial task to allocate testing and code-review resources to areas that require the most attention. To overcome the above problem, researchers have developed software bug prediction models that can help practitioners to predict the most bug-prone software entities. Although, software bug prediction is a very popular research area, yet its industrial adoption remains limited.
In this thesis, we investigate three possible issues with the current state-of-the-art in software bug prediction that affect the practical usability of prediction models. First, we argue that current bug prediction models implicitly assume that all bugs are the same without taking into consideration their impact. We study the impact of bugs in terms of experience of the developers required to fix them. Second, only few studies investigate the impact of specific type of bugs. Therefore, we characterize a severe type of bug called Blocking bugs, and provide approaches to predict them early on. Third, false-negative files are buggy files that bug prediction models incorrectly as non-buggy files. We argue that a large number of false-negative files makes bug prediction models less attractive for developers. In our thesis, we quantify the extent of false-negative files, and manually inspect them in order to better understand their nature.
Computing and Information Sciences (Ph.D.)
Department, Program, or Center
PhD Program in Computing and Information Sciences
Valdivia-Garcia, Harold, "Understanding the Impact of Diversity in Software Bugs on Bug Prediction Models" (2016). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus