Deep Learning vs. Traditional Machine Learning for Software Defect Prediction: A Meta-Analysis
Year: 2026
Type: review
Study designs: Systematic Review, Meta-Analysis
Abstract: **Deep Learning vs. Traditional Machine Learning for Software Defect Prediction: A Meta-Analysis** **Weixiang Gan** * Post-Doctorate Researcher, Lincoln University College, 47301 Petaling Jaya, Malaysia **Jialin Liu** School of Management and Marketing, Faculty of Business and Law, Taylor's University, No. 1, Jalan Taylor’s, 47500 Subang Jaya, Malaysia **Mengfei Xiao** Post-Doctorate Researcher, Lincoln University College, 47301 Petaling Jaya, Malaysia **Tara Ahmed Mohammed** Graduate School of Business, SEGi University, Kota Damansara, 47810 Petaling Jaya, Malaysia **Abstract:** Deep learning (DL) is now routine in software defect prediction (SDP), yet how much it improves on traditional machine learning (ML), how stable that improvement is, and what governs it remain contested. We synthesized 45 empirical studies published between 2015 and 2024, comprising 1540 performance estimates, using Hedges’ _g_ of the area underx the receiver operating characteristic curve (AUC) and a random-effects model. The pooled effect was _g_ = 0.61 (95% CI 0.53–0.70; _p_ < 0.001), but heterogeneity was substantial (_I_2 = 87.7%; _τ_2 = 0.076) and the 95% prediction interval was 0.07–1.16. The lower pre-diction limit is close to the null and is therefore more decision-relevant than the positive mean alone: a new setting may show little practical gain. Hybrid architectures produced the largest subgroup estimate (_g_ = 0.90), whereas the gain under cross-project defect prediction was less than half that under within-project defect prediction (0.31 vs. 0.67). A mild publication-year association explained only about 15% of be-tween-study heterogeneity, and no sample-size association was detected. Publication-bias and leave-one-out diagnostics did not identify a single dominant study, but these diagnostics do not eliminate bias from primary-study design or baseline tuning. The evidence therefore supports a conditional, not universal, DL advantage whose practical value depends on architecture, validation protocol, data distri-bution, tuning quality, and deployment cost. **Keywords:** Software Defect Prediction, Deep Learning, Machine Learning, Meta-Analysis, Effect Size, Heterogeneity, Publication Bias **Citation:** W. Gan, J. Liu, M. Xiao, and T. A. Mohammed, “Deep learning vs. traditional machine learning for software defect prediction: A meta-analysis,” Engineering and Computing Insights, pp. 3–21, 2026, doi: 10.67728/eci.2026.002.
Institutions: Lincoln University College, Taylor's University, SEGi University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7204137930