"Feature Selection for Software Fault Prediction Using Machine Learning: A Systematic Evidence-Based Review of Techniques, Benchmarks, and Emerging Trends"

This review has systematically examined the effects of feature selection on software fault prediction across 25 primary studies published between 2022 and 2026. The evidence strongly supports the utility of FS as a preprocessing step across filter, wrapper, embedded, hybrid, and ensemble paradigms, appropriately selected feature subsets consistently improve predictive accuracy, reduce training time, and enhance model interpretability compared to full-feature baselines. No single FS paradigm dominates in all contexts; rather, the optimal choice depends on dataset characteristics, computational constraints, and the degree to which interpretability is prioritized. Filter methods offer efficiency and simplicity but may underperform on datasets with complex feature interactions or severe class imbalance. Wrapper methods deliver the highest accuracy but carry computational costs that constrain their application to large-scale industrial repositories. Embedded methods such as Random Forest importance and LASSO provide a practical balance, and their integration with SHAP-based explainability is an especially promising direction. Hybrid and ensemble FS strategies emerge as robust alternatives that improve selection stability and cross-project transferability. The field is entering a transformative phase shaped by the adoption of deep learning and LLMs for feature representation. Traditional handcrafted metrics face competition from learned embeddings, yet also remain valuable for their interpretability and low data requirements. The future of FS in SFP likely lies in hybrid pipelines that judiciously combine metric-based and learned features, guided by causal understanding and evaluated under rigorous, reproducible experimental protocols. Addressing the methodological gaps identified in this review will be essential to translating research advances into reliable, deployable fault prediction tools for software engineering practice.

[1]. J. Goyal, & B. Kishan, “Progress on Machine Learning Techniques for Software Fault Prediction”, International Journal of Advanced Trends in Computer Science and Engineering, 2019, volume 8, Number 2, page No. 305-311, ISSN 2278-3091, https://doi.org/10.30534/ijatcse/2019/33822019 [2] Hosseini, S., Turhan, B., & Gunarathna, D. (2022). Mutual information-based feature selection for within-project and cross-project software defect prediction. Journal of Systems and Software, 185, 111155. https://doi.org/10.1016/j.jss.2021.111155 [3]. Jyoti Goyal and Bal Kishan, “Empirical analysis of object-oriented metrics for prediction of Software faults”, Bioscience Biotechnology Research Communications, 2020, Special Issue Volume 13, Number 15, Page No. 292- 296. [4]. Ghotra, B., McIntosh, S., & Hassan, A. E. (2022). A large-scale study of the impact of feature selection on software defect prediction. IEEE Transactions on Software Engineering, 49(3), 1207–1228. https://doi.org/10.1109/TSE.2022.3165289 [5]. Kaur, A., & Pallavi, S. (2023). ReliefF-based feature selection for software fault prediction: A comparative evaluation on Eclipse and PROMISE datasets. Expert Systems with Applications, 213, 118897. https://doi.org/10.1016/j.eswa.2022.118897 [6]. Ni, C., Xia, X., Lo, D., & Hassan, A. E. (2022). Just-in-time defect prediction on JavaScript projects: A replication study. Empirical Software Engineering, 27(3), 59. https://doi.org/10.1007/s10664-021-10086-4 [7]. Fan, G., Diao, X., Yu, H., Yang, K., & Chen, L. (2019, December). Deep semantic feature learning with embedded static metrics for software defect prediction. In 2019 26th Asia-Pacific Software Engineering Conference (APSEC) (pp. 244-251). IEEE. [8]. Muhamad, F. P. B., Siahaan, D. O., & Fatichah, C. (2018). Software fault prediction using filtering feature selection in cluster-based classification. IPTEK Journal of Proceedings Series, (1), 59-64. [9]. Malhotra, R., Chawla, S., & Sharma, A. (2025). Software defect prediction based on multi-filter wrapper feature selection and deep neural network with attention mechanism. Neural Computing and Applications, 37(27), 22621-22648. [10]. Balogun, A. O., Basri, S., Capretz, L. F., Mahamad, S., Imam, A. A., Almomani, M. A., ... & Kumar, G. (2021). Software defect prediction using wrapper feature selection based on dynamic re-ranking strategy. Symmetry, 13(11), 2166.