Hybrid balancing method for software fault prediction leveraging stacked ensembles

Hybrid balancing method for software fault prediction leveraging stacked ensembles

연구 분야: Verification

학회: International Journal of Information Technology

Early Fault Prediction identifies the fault-prone modules beforehand and saves the testing resources. It facilitates the timely delivery of quality products with reduced costs. High traction of machine learning (ML) based Software Fault Prediction (SFP) models is pronounced but challenged by Class Imbalance (CI) condition of the defect datasets. This study presents a promising SFP model named UOB_STACK with innate potential to deal with CI utilizing data balancing and stacked ensemble. Eight datasets are sourced from NASA and PROMISE repositories for the experimentation. UOB_STACK improves the base models by 32.2% and outperforms the selected baselines by 15.9%. The results are statistically validated with p-statistic 0.0078 for the Wilcoxon Signed Rank-Sum Test with post hoc Bonferroni Test at α = 0.05.