Assessing the success of agile software projects using machine learning techniques: A case study

Agile methodologies, particularly Scrum, enable software development teams to manage projects more effectively by offering flexibility, collaboration, and small, functional deliveries. However, the lack of systematic frameworks addressing the factors influencing sprint success prevents teams from receiving clear guidance to improve their performance. This thesis aims to evaluate sprint success in Agile software development processes and provide recommendations to enhance these processes. In this study, factors affecting sprint success were examined using industrial projects from Siemens and a publicly available dataset. Machine learning models were employed to enhance the accuracy of sprint planning and to support teams' decision-making processes with data-driven predictions. Additionally, explainability techniques such as SHAP and LIME were applied to increase the transparency of machine learning models, often referred to as "black-box" systems. The findings of the research identify key elements contributing to sprint success and provide actionable insights to help teams improve their processes. Moreover, the results demonstrate that data-driven approaches offer more reliable outcomes compared to traditional subjective estimation methods, helping teams achieve sprint goals more effectively. Finally, it is concluded that explainable machine learning models enable software development teams to understand better and trust prediction results, fostering confidence in the use of these tools.