Hybrid Approach to Software Defect Prediction Using Stacking Ensemble

Abstract

Software defect prediction plays a critical role in improving software quality and reducing maintenance costs by identifying fault-prone modules early in the development lifecycle. Traditional machine learning models often suffer fromlimited generalization capability when applied to diverse and imbalanced software datasets. To address these challenges, this study proposes a hybrid approach for software defect prediction using a stacking ensemble technique.
The proposed method integrates multiple base learners—such as Decision Trees, Random Forest, Support Vector Machines, and Gradient Boosting—at the first level to capture diverse patterns in software metrics. Their predictions are then combined using a meta-learner, typically Logistic Regression or XGBoost, which learns to optimally aggregate the outputs of base models.