BUBBLE: A Scalable and Efficient Bellwether Discovery Method for Large-Scale Software Engineering Datasets

Bellwether analysis is a widely-used technique in software engineering for identifying representative projects that can be used to predict outcomes in other projects. However, existing methods, such as the one proposed by Krishna et al., require extensive comparisons, making them inefficient for large datasets. In this paper, I present a novel approach called BUBBLE, a hierarchical bellwether discovery method designed to scale with large datasets. BUBBLE clusters projects into hierarchical levels, reducing the number of comparisons by a factor of mmm, where mmm is the number of clusters at the leaf level. This method not only improves performance but also maintains prediction quality. Experimental results on a defect prediction dataset of 697 projects show that BUBBLE reduces bellwether discovery time from hours to mere minutes while yielding high recall, precision, and defect prediction metrics. I also discuss threats to validity and demonstrate that BUBBLE is both statistically and empirically effective for large-scale bellwether analysis in software engineering.