Software Process Extraction and Identification

Industrial software is planned, managed, and created using a variety of approaches: sometimes a formal Software Development Life-cycle (SDLC) model is used, sometimes an approach that is less formal but has clearly identifiable stages is employed, and sometimes little if any discernible process is followed. In this paper, we extract and correlate the software development process with behaviour and data found within a project’s source control repository. We do so by analyzing fine grained changes, revisions, and aggregations of revisions, so that we can correlate them with the stage of the software development process that the project was in at the time they were made. To label intervals of revisions we use machine learning and artificial intelligence techniques including N-Nearest Neighbours classifiers, Markov models, Hidden Markov Models, Markov Decision Processes and Partially Observable Markov Decision Processes. These techniques initially learn the stages from annotated data and then classify unknown data. We describe how to pose the problem using these tools and we evaluate their effectiveness on several case studies.