Replication package for Auditing Class Rebalancing and Decision Thresholds in Software Defect Prediction under Feature-Group Separation

Files

1.zip corresponds to Online Resource 1. It contains scientific source-code snapshots, experimental protocols, aggregate results, provenance records and bounded verification scripts. The materials cover the original comparisons, record-budget analyses, evaluation-mask and training-only sensitivity analyses, alternative threshold objectives, a qualified DNC reconstruction and an unsuccessful policy-gate development experiment.

2.zip corresponds to Online Resource 2. It contains project-level inputs, all 72 leave-one-project-out influence estimates, summary results, the analysis plan, execution records and a standalone numerical verifier. 

Scope and limitations

The packages support numerical reconstruction of saved aggregate results and inspection of the included implementations. They do not provide a complete end-to-end model-retraining archive. Third-party raw feature tables, record-level labels and predictions, and fitted models are not included. 

The 18 Apache projects share a common data lineage and are not treated as 18 independent organizations. 

Rights

The CC BY 4.0 license applies to the author's original materials. Third-party materials retain their original terms.