Improving Software Fault Prediction With a Hybrid DE-WOA Optimizer and ANFIS-Enhanced Ensemble Learning

Software Fault Prediction (SFP) plays a critical role in improving software quality by identifying defect-prone modules early in the development lifecycle. However, existing SFP approaches often struggle to balance predictive performance, interpretability, and robustness, particularly when faced with high-dimensional, imbalanced datasets and correlated features. This paper proposes a novel hybrid methodology that integrates a Differential Evolution–Whale Optimization Algorithm (DE-WOA) for efficient feature selection, an Adaptive Neuro-Fuzzy Inference System (ANFIS)-enhanced ensemble for dynamic prediction, and explainability techniques (SHAP, permutation importance, PCA) for robust model interpretation. The proposed framework is evaluated on the PROMISE Ant 1.3 dataset where it reduces dimensionality (selecting 2 features for XGBoost and 4 for Logistic Regression), improves Logistic Regression ensemble accuracy to 0.84 (F1-score 0.817, AUC-ROC 0.714), and outperforms the XGBoost-based ensemble in both predictive performance and interpretability. Statistical analysis using the Wilcoxon test (p-value = 0.1408) suggests that the models are not significantly different in performance, yet practical improvements in non-defective recall and feature interpretability are notable. This unified framework addresses key challenges in SFP by combining metaheuristic feature optimization, ensemble learning, and explainable AI, offering a scalable and interpretable solution for software quality assurance.