Graph dataset of OOD metrics from UML class diagrams

A graph-based dataset for software defect prediction, constructed from UML class diagrams of 70 versioned open-source Java projects across 5 bug repositories (BugCatchers, BugPrediction, GitHub, PROMISE, Zimmerman). Each graph represents one project version: nodes are classes or interfaces with 66 object-oriented design (OOD) metrics as features; edges represent UML structural relationships (generalization, realization, association, dependency). Defect labels are derived from the Unified Bug Dataset (Ferenc et al., 2018). Class diagrams were reverse-engineered from Java source code using Enterprise Architect and exported in XMI format. Metrics were computed using a custom extraction pipeline available at the companion GitHub repository.