Enhanced Network Intrusion Detection via Gradient Boosting Tuned by Emperor Penguin Optimization Algorithm (EPOA)

Abstract

Protecting network infrastructures from increasingly complex cyberthreats requires the use of intrusion detection systems, or IDSs. However, because of changing attack patterns and high data dimensionality, it is still difficult to differentiate between malicious and benign network activity. In order to improve IDS performance, this study critically evaluates six popular machine learning classifiers: Random Forest (RF), Gradient Boosting (GB), Decision Tree (DT), XGBoost (XGB), AdaBoost (AB), and K-Nearest Neighbors (KNN). Two sophisticated hyperparameter tuning methods, Grid Search (GS) and the Emperor Penguin Optimization Algorithm (EPOA), were used to increase predictive accuracy and model robustness. With accuracy, precision, recall, and F1-score values of 0.9997, 0.9898, 0.9999, and 0.9948, respectively, the optimized Gradient Boosting (EPOA-GB) model outperformed the others. Important contributing features were also found using SHAP-based interpretability analysis, which provided insightful information about the classification procedure. The models became more scalable for deployment when Principal Component Analysis (PCA) was used to reduce dimensionality, improving generalization and computational efficiency. These results show how well ensemble classifiers and intelligent optimization work together to reduce false alarms, a crucial requirement for real-time intrusion detection. This work provides practical guidelines for implementing high-performance IDSs and highlights the importance of future validation across diverse datasets and deployment environments to ensure robustness and adaptability in real-world cybersecurity scenarios.