Comparison of Machine Learning Models in Detecting Various Types of DDoS Attacks

Abstract

Distributed Denial of Service attacks remain a major network-security threat because they can overwhelm computing resources, disrupt digital services, and cause substantial operational losses. The emergence of Generative Artificial Intelligence has further increased the urgency of adaptive and efficient traffic-based detection because intelligent systems may support defensive analysis while also enabling increasingly automated attack strategies. This study compared Random Forest, Multilayer Perceptron Classifier, and Extreme Gradient Boosting for detecting DDoS traffic using the CICDDoS2019 dataset. The dataset was processed through duplicate removal, missing-value checking, feature filtering, label encoding, feature scaling, and data splitting. The models were evaluated using accuracy, precision, recall, F1-score, Receiver Operating Characteristic Area Under the Curve, and cross-validation score. Random Forest achieved the highest accuracy of 0.992770 and cross-validation score of 0.992684, whereas Extreme Gradient Boosting achieved the highest ROC AUC of 0.993765. The results demonstrate that Random Forest provided the most stable overall performance, while Extreme Gradient Boosting offered the strongest class-discrimination capability.