Fuzzy Clustering of Open-Source Software Quality Data: A Case Study of Mozilla

We present a fuzzy cluster analysis of software quality data extracted from the Mozilla open-source Web browser. This is a new dataset that combines object-oriented software quality metrics with the number of defects per code unit. We undertake a fuzzy cluster analysis of this dataset, which for the first time addresses the use of both hyperspherical and hyperellipsoidal fuzzy clusters (using the Gath-Geva algorithm) in software quality analysis. Using a Pareto analysis based on the fuzzy clusters, we were able to identify groups of modules having higher defect densities than would be found by merely ranking modules based on any single software metric.