An Explainable Feature Selection and Stacking Ensemble Framework for Software Fault Prediction

Abstract

The increasing complexity of contemporary software systems makes accurate and interpretable software fault prediction a challenging task. Although ensemble learning techniques have demonstrated promising predictive performance, existing approaches continue to face challenges related to feature selection, cross-project consistency, and model interpretability. This study proposes a Hybrid Boosted Stacking Ensemble Framework (HBSEF) for software fault prediction that integrates hybrid feature selection, ensemble learning, statistical validation, and SHAP-based interpretability. The proposed framework incorporates data preprocessing, class-imbalance handling using Random Oversampling, and a hybrid feature selection strategy integrating filter, embedded, and wrapper methods to identify software metrics with consistent predictive importance across multiple software projects. The core predictive model employs CatBoost, XGBoost, and LightGBM as base learners, with XGBoost serving as the meta-learner in the stacking ensemble. The framework is evaluated on an unseen test set using standard classification metrics, while paired t-testing is employed to assess the statistical significance of performance differences. Experimental results identify LOC, CBO, WMC, RFC, LCOM, and MAX_CC as the most influential software metrics. The proposed framework demonstrates statistically significant improvements over the baseline ensemble models, as validated through paired t-testing (t = 4.53, p = 0.0004), indicating the effectiveness of the proposed approach in enhancing software fault prediction performance. SHAP-based explanations provide both global and local interpretations, revealing the contribution of individual software metrics and base-model predictions to fault classification. Overall, the findings indicate that integrating principled feature selection with a boosting-based stacking ensemble can improve software fault prediction performance while providing greater transparency for software quality management.