Hybrid Firefly and Particle Swarm Optimization for parameter tuning of XGBoost: Network Intrusion Detection

Abstract This paper presents an intrusion detection framework combining machine learning, feature engineering, and metaheuristic optimization to enhance network security. The approach employs Extreme Gradient Boosting (XGBoost) for automatic feature selection and model evaluation using confusion matrix-based metrics, including accuracy, precision, recall, F1- score, true positive, and false positive rates. To optimize hyperparameters in high-dimensional datasets, a novel hybrid Firefly Algorithm–Particle Swarm Optimization (FA–PSO) is introduced, improving the exploration of optimal configurations. The framework incorporates comprehensive data preprocessing, feature engineering, and a stacked ensemble learning strategy for improved detection accuracy and generalization. Evaluations on the NSL-KDD dataset demonstrate superior performance over traditional models, confirming the effectiveness of integrating feature engineering with hybrid optimization for scalable, real-time intrusion detection.