Detection of DDoS Attack Using Machine Learning Models

Distributed Denial of Service (DDoS) attacks remain one of the fastest-moving threats in cybersecurity, attacking important network infrastructure through flooding systems with malicious traffic and thereby degrading legitimate services. This paper envisions an end-to-end framework using machine learning (ML) for enabling precise and effective DDoS attack detection, using the CICIDS2017 benchmark dataset provided by the Canadian Institute for Cybersecurity. We conduct a systematic comparison of six supervised learning algorithms-K Nearest Neighbor (KNN), Support Vector Machine (SVM), Neural Net- works (NN), Random Forest, CatBoost, and XG Boost-with the aim of determining the topperforming model to discriminate between benign and malicious traffic patterns. Our approach is centered around rigorous data preprocessing, feature engineering, and hyperparameter tuning for ensuring generalizability and bypassing the challenge of overfitting. Experimental results indicate that ensemble-based approaches, i.e., Random Forest, CatBoost, and XGBoost, achieve near perfect classification accuracy (99.9%) with enhanced precision, recall, and F1-scores relative to conventional models like KNN and SVM. In addition, we assess computational efficiency and interpretability of the models to support practical deployment within real-world network environments. This paper not only confirms the potency of advanced ensemble methodologies in the detection of DDoS attacks but also presents a scalable, data-driven framework that can be modified to new emerging attack vectors. Future extensions include improvements in real time detection performance and adversarial robustness testing to combat zeroday attacks.