Local versus global models for effort-aware defect prediction
Bibliographic record
Abstract
Software entities (e.g., files or classes) do not have the same density of defects and therefore do not require the same amount of effort for inspection. With limited resources, it is critical to reveal as many defects as possible. To satisfy such need, effort-aware defect prediction models have been proposed. However, the performance of prediction models is commonly affected by a large amount of possible variability in the training data. Prior studies have inspected whether using a subset of the original training data (i.e., local models) could improve the performance of prediction models in the context of defect prediction and effort estimation in comparison with global models (i.e., trained on the whole dataset). However, no consensus has been reached and the comparison has not been performed in the context of effort-aware defect prediction. In this study, we compare local and global effort-aware defect prediction models using 15 projects from the widely used AEEEM and PROMISE datasets. We observe that although there is at least one local model that can outperform the global model, there always exists another local model that performs very poorly in all the projects. We further find that the poor performing local model is built on the subset of the training set with a low ratio of defective entities. By excluding such subset of the training set and building a local effort-aware model with the remaining training set, the local model usually underperforms the global model in 11 out of the 15 studied projects. A close inspection on the failure of local effort-aware models reveals that the major challenge comes from defective entities with small size (i.e., few lines of code), as such entities tend to be correctly predicted by the global model but missed by the local model. Further work should pay special attention to the small but defective entities.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.008 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.003 | 0.002 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".