Improving change prediction with fine-grained source code mining

Abstract

The thesis proposes a software-change prediction approach that is based on mining fine-grained evolutionary couplings from source code repositories. Here, fine-grain refers to identifying couplings between source code entities such as methods, control structures, or even comments. This differs from current source code mining techniques that typically only identify couplings between files or fairly high-level entities. Furthermore, the model combines the mined evolutionary couplings with the estimated changes identified by traditional impact analysis techniques (e.g., static analysis of call and program-dependency graphs). The research hypothesis is that software-change prediction using the proposed synergistic approach results in an overall improved expressiveness (i.e., granularity and context given to a developer) and effectiveness (i.e., accuracy of the prediction).