Feature-Based Characterization of GitHub OSS Projects for Performance Assessment
José Manuel Sánchez Ruiz, Guillermo Ciria González, Miguel Ángel Olivero, Francisco José Domínguez Mayo, David Benavides · 2025
GitHub hosts a huge variety of Open-Source Software projects, each one with a particular project configuration. Such a diversity, and the freedom on the features combination, limits the ability to systematically compare them. Benchmarking, discovering best-practices, or informed decision-making is hindered due to the absence of structured characterization. This study captures the variabilities and commonalities among open-source projects, with the goal of identifying configurations that correlate with a higher performance. To deal with this wide diversity we outline a structured feature-based model considering the Software Product Line paradigm. By aligning our approach with SPL modeling, we can correlate projects’ configurations with their performance metrics, allowing the assessment and comparison of projects’ efficiency. To build and improve this model, we manually developed an initial feature model based on the analysis of GitHub features. We then used Large Language Models to expand and improve the model. We applied structured prompts to systematically explore new feature candidates across five models. Then, we analyzed the performance of 40 well-known OSS projects with technical relevance. The results helped enrich the feature model by linking specific GitHub configurations to performance indicators across these projects. This structured characterization forms the basis for a recommendation system designed to improve OSS development practices. By using feature-based analysis, our work supports better benchmarking, decision-making, and performance optimization in OSS ecosystems from a Software Product Line perspective.