DDOS ASSAULT DETECTION-MITIGATION: AMALGAMATION OF GREY WOLF OPTIMIZER-SALP SWARM ALGORITHM, HYBRID MODEL AND DEEP Q-NETWORK

Abstract

Traditional DDoS assault detection depend on static signatures, manually built features, individual classifiers, or hard-coded mitigation rules, and hence ineffective due to concept drift, zero-day attack patterns and bursts of legitimate/malicious traffic. In the proposed intelligent DDoS assault detection framework, adaptive feature selection, ensemble detection, multi-factor threat intelligence and deep reinforcement learning for the optimization of the response mechanism are integrated in a closed loop. It incorporates feature reduction techniques, Grey Wolf Optimizer and Salp Swarm Algorithm, adaptive classification techniques, Random Forest, Support Vector Machine and XGBoost, anomaly aware threat scoring for contextual risk assessment, and a Deep Q-Network for dynamic learning of mitigation policies. A continual learning layer is added on to facilitate feedback-driven retraining and survivability in the presence of the evolving assault conditions. The proposed framework with strongest configuration, achieved 98% accuracy and 0.9944 AUC with 0.000339 false alarm rate and 0.000326 false discovery rate.