Retrospective Project Analysis Using the Expectation-Maximization Clustering Algorithm

Schedule slips are often the reason for failed projects or low-quality software. Therefore, investigation if a project was on schedule is an important task when analyzing software projects in retrospective. In this paper, we present a data-driven approach for the retrospective determination of project phases through a clustering algorithm. The analysis is based on software metrics measured at different points of time during the project execution. We will describe how the data can be collected, prepared and analyzed. Our findings are validated through a case study where we analyzed two large scale open-source projects. The results show that it is possible to successfully identify the final phase of a project using our approach.