Stratified Meta-Learning for DDoS Attack Classification and Counteraction using Rate Limiting

Network infrastructure encounters severe threats from Distributed Denial of Service (DDoS) attacks that become more complex and occur more frequently, resulting in service interruptions together with financial burdens and compromised cybersecurity. Traditional IDS operating independently with machine learning together with deep learning faces performance limitations while dealing with false alarms along with insufficient adaptation to changing attack patterns and compromising effective real-time processing. This research presents the Stratified Meta-Learning framework for DDoS Attack Classification and Counteraction Using Rate Limiting which integrates superior stacked ensemble learning with advanced intelligent realtime attack mitigation. The proposed approach implements Extra Trees Classifier (ETC) together with Feedforward Neural Network (FNN) as base learners to discover various feature correlations and network traffic data sequences. The base model predictions pass through CatBoost as the meta-model allowing the system to discover complex associations to boost its prediction accuracy. Training and evaluation of the model happens with the help of the UNSWNB15 dataset which provides multiple attack scenarios for comprehensive benchmark testing of the model across different cyber threats. The research development introduces Token Bucket Algorithm as a Rate Limiting system to actively control network traffic speed which defends against DDoS attacks in current time. Our stratified meta-learning framework outperforms traditional IDS systems through extensive tests which yield ≥96% accuracy and better precision-recall performance while decreasing false positive detections.