Publications

Abstract

Distributed denial of service (DDoS) outbreaks are growing as a serious cybersecurity concern. They can disable online services in minutes by flooding network resources. Traditional rule-based systems have a hard time detecting new or zero-day attacks because they depend on static signatures. Existing emerging models often perform poorly due to bad hyperparameter tuning. In this research study a hybrid framework that collaborates Particle Swarm Optimization (PSO) with a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) architecture for real-time DDoS detection and response. PSO minimize the error in hyperparameters of the CNN-LSTM model, which reduces the need for manual tuning and enhances adaptability to different traffic situations. The CNN part effectively captures spatial traffic patterns, while the LSTM models temporal relationships, allowing for thorough traffic analysis. Experiments using the CICDDoS2019 dataset and synthetic imbalanced datasets show impressive results. The proposed system achieves 98.83% accuracy, 98.84% precision, 98.83% recall, and 98.83% F1-score. A comparison with optimization-based methods, such as Genetic Algorithm (GA) and Grey Wolf Optimizer (GWO) confirms the success of this approach. The framework also works with an automated response system that uses firewall-based blocking to ensure quick reactions. The results underline its scalability, adaptability, and reliability for enterprise and cloud network security.