A Machine Learning -Based Classification and Prediction Technique For DDOS

Abstract

Modern networks continue to experience rising Distributed Denial of Service
(DDoS) attacks that are increasingly designed to evade traditional detection systems. Manual
approaches such as monitoring traffic logs, identifying abnormal spikes, and blocking suspicious
IP ranges through firewall rules are slow and error-prone. These methods become ineffective
under high volume traffic, cannot adapt to evolving attack patterns, and often fail to detect subtle
or complex behaviors. This project explores the use of Support Vector Machines, Random Forest
based models, and particularly XGBoost as a more reliable approach for classifying and
predicting DDoS attack behavior using a recent, high-volume dataset. Several preprocessing
steps were applied to ensure meaningful feature representation, and XGBoost was chosen for its
ability to handle sparse data, complex patterns, and noisy samples with strong generalization
capability. Performance evaluation using accuracy, precision, recall, and confusion-matrix
analysis shows that XGBoost significantly improves the detection of subtle attack patterns
compared to earlier techniques. The findings demonstrate strong potential for deploying this
model in real-time environments and highlight gradient-boosting methods as a promising
direction for future automated threat-detection research.