Mitigating Low Bus Factor Risks:A Proactive Approach to Address Single Points of Failure in Software Development Teams

Abstract

Abstract This research addresses a common problem in software engineering where the knowledge is concentrated in a small subset of developers within a team which can lead to serious problems on their departure. The bus factor is a vital metric for assessing the resilience of software projects by estimating the risks associated with the loss of key contributors. Traditional approaches often rely on simplistic heuristics, failing to capture the complex collaboration dynamics and distributed knowledge within development teams. This thesis presents a graph-based method that models contributor interactions through social network analysis to identify critical knowledge holders. By incorporating multiple dimensions—such as lines of code contributions, file ownership diversity, interaction centrality, and knowledge decay—the proposed approach offers a more comprehensive and context-aware assessment of project risk. To enhance accuracy, the method integrates project context data from Jira issues, enabling the identification of essential contributors who may not write significant amounts of code but play key roles in the development process. Additionally, the tool supports efficient onboarding through automated documentation generation, facilitating smoother knowledge transfer. This study adopted a pragmatic research philosophy with an inductive, mixed-methods approach, leveraging Design Science Research Methodology (DSRM) to develop and evaluate a system for identifying critical knowledge holders in software projects. Through iterative development, the tool was enhanced across three stages: analyzing version control data to detect high-risk areas, integrating issue-tracking data for contextual insights, and generating detailed documentation to support knowledge transfer. The outcomes were validated through comparative analysis and stakeholder feedback, ensuring both technical accuracy and real-world applicability. The solution is delivered as an interactive, web-based application optimized for usability and visual analysis while evaluating its effectiveness using GitHub repositories, beginning with university projects and extending to 12 real-world industry projects from diverse organizations. The tool successfully identified all key contributors in these 12 projects with an overall accuracy of 85.56% , demonstrating its robustness across different team sizes, repository structures, and workflows. Compared to traditional techniques, this approach provides deeper insights for proactive risk mitigation and significantly strengthens team resilience.