Learning from bug-introducing changes to prevent fault prone code

A version control system, such as CVS/SVN, can provide the history of software changes performed during the evolution of a software project. Among all the changes performed there are some which cause the introduction of bugs, often resolved later with other changes. In this paper we use a technique to identify bug-introducing changes to train a model that can be used to predict if a new change may introduces or not a bug. We represent software changes as elements of a n-dimensional vector space of terms coordinates extracted from source code snapshots. The evaluation of various learning algorithms on a set of open source projects looks very promising, in particular for KNN (K-Nearest Neighbor algorithm) where a significant tradeoff between precision and recall has been obtained. © 2007 ACM