The rapid escalation of sophisticated cyberattacks demands intrusion detection systems that not only achieve high accuracy but also provide transparency and fairness in decision-making, addressing critical gaps in current machine learning-based cybersecurity solutions. This study proposes a trustworthy intrusion detection framework that integrates Bald Eagle Search (BES) optimization with the XGBoost classifier and SHAP-based explainability to enhance both predictive performance and interpretability. The BES algorithm efficiently fine-tunes key XGBoost hyperparameters to overcome limitations of traditional models that struggle with complex, high-dimensional intrusion patterns. Evaluated on a real-world cybersecurity dataset, the BES-optimized XGBoost model achieves a substantially improved accuracy of 99.7%, outperforming baseline XGBoost (87.5%) and other meta-heuristic optimizers including GA, ACO, and PSO. SHAP visualizations further reveal feature-level contributions, ensuring transparent and accountable detection decisions, while fairness analysis highlights disparities across browser types, promoting responsible and bias-aware cybersecurity deployment. Overall, the proposed model offers a robust, interpretable, and ethically aligned solution for detecting modern cyber threats, contributing significantly to the advancement of transparent and high-performance intrusion detection systems.
Kundu et al. (Tue,) studied this question.