Boosting Intrusion Detection Against DDoS Attacks Using a Feature Engineering-Based Fine-Tuned XGBoost Model

Network security is seriously threatened by distributed denial-of-service (DDoS) attacks, which calls for sophisticated intrusion detection systems that can rapidly identify and mitigate such threats. Despite their widespread use in intrusion detection against DDoS attacks, machine learning methods still suffer accuracy degradation due to inadequate data pre-processing and computational inefficiency. This study combined a fine-tuned extreme gradient boosting (XGBoost) model with correlation-based feature selection—for efficient feature selection—to effectively maximize detection accuracy while lowering computing overhead. Both correlation-based feature selection and XGBoost contribute to boosting the final model's efficiency. To evaluate the proposed model, different metrics were employed over three DDoS data sets, considering both binary and multi-classification scenarios. Experimental findings demonstrate that the proposed XGBoost achieves highly competitive accuracy. For the Network Security Laboratory–Knowledge Discovery Databases data set, University of New South Wales–Network Behavior15 data set, and Canadian Institute for Cybersecurity–Intrusion Detection System–2017 data set, the model secures 0.995, 1.000, and 0.999 and 0.996, 0.885, 0.998 for binary and multi-classification, respectively, outperforming its rival models.