Enhancing Software Defect Prediction:Integrating SAINT with Ant Colony Optimization and Explainable AI

Software Defect Prediction (SDP) enhances software quality by identifying error-prone modules early in development. While deep learning models like SAINT improve prediction accuracy, their complexity limits interpretability. This study evaluates SAINT against traditional models (XGBoost, Random Forest, CatBoost) and addresses interpretability using Ant Colony Optimization (ACO) for feature selection and hyperparameter tuning, along with LIME for explaining predictions. Performance metrics, including Prediction Detection (PD), Prediction False Alarm (PF), Balance, and Fault Identification Rate (FIR), demonstrate that SAINT outperforms traditional models in accuracy. The integration of SAINT, ACO, and LIME provides a robust, transparent framework for real-world SDP applications.