What the study found
The review found that Support Vector Machine and Random Forest were the strongest methods in the studies it examined, based on accuracy, precision, recall, and F1-score. It also found that public datasets, especially from the PROMISE and NASA Metric Data Program repositories, are widely used in software fault prediction.
Why the authors say this matters
The authors say the review is meant to advance research in software fault prediction and support the development of high-quality software products by improving defect predictability. The study also suggests it will be useful to researchers by providing an updated overview of the literature.
What the researchers tested
The paper reviewed 45 articles published between 2023 and 2025 from the IEEE, Springer, ACM, and ScienceDirect digital libraries. It covered factors influencing software fault prediction, prediction techniques, datasets and software metrics, evaluation metrics, model selection criteria, and challenges in current solutions.
What worked and what didn't
Across the reviewed articles, Support Vector Machine and Random Forest had the best reported performance on the metrics named in the abstract. The use of public datasets, particularly PROMISE and NASA Metric Data Program product-metric datasets, was associated with improved model performance. The abstract does not identify which approaches performed poorly overall.
What to keep in mind
The summary is based only on a review of 45 articles and the findings reported in their abstracts and selected results. The abstract does not provide detailed limitations, and it does not specify the full range of models or datasets beyond those named.
Key points
- The review covered 45 articles on software fault prediction published from 2023 to 2025.
- Support Vector Machine and Random Forest were reported as the best-performing methods in the reviewed studies.
- Public datasets from PROMISE and NASA Metric Data Program were widely used and linked to improved model performance.
- The review examined prediction techniques, datasets, software metrics, evaluation metrics, model selection criteria, and challenges.
- The authors say the review may help advance research and improve defect predictability.
Disclosure
- Research title:
- Support Vector Machine and Random Forest led software fault prediction
- Authors:
- Ruchika Aggarwal, Kamaljit Kaur
- Publication date:
- 2026-01-28
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.