Analysis of DDoS attack and defense system in cloud environment

Abstract As cloud computing continues to govern digital infrastructure, ensuring the security and availability of cloud services is increasingly important. One of the most disruptive attacks on cloud settings is the Distributed Denial of Service (DDoS) attack, which overloads cloud servers with fraudulent traffic, causing service outages and draining resources. Traditional DDoS detection techniques frequently fail to adapt to the dynamic and large-scale nature of cloud traffic. Previous solutions either fail to generalize to different threat patterns or lack smart decision-making capabilities for real-time protection. By using advanced feature selection and ensemble classification, the system improves detection accuracy and timeliness. The emphasis is on combining effective feature selection with ensemble learning to enhance prediction accuracy. This paper presents an innovative algorithm, EnhancedCloudDDoS-Guard, that incorporates a Multi-Stage Intelligent Feature Selector (MSIFS) combining Mutual Information, Chi-Square, and Recursive Feature Elimination (RFE) with majority voting to select the most pertinent features. The selected features are used to train EnsembleNet, an ensemble classifier that combines LightGBM, CatBoost, and ExtraTreesClassifier with weighted soft voting. The model is trained and tested on a labeled cloud traffic dataset that includes attributes such as request rate, CPU usage, blacklist status, and port access behavior. Experimental assessment on a dataset of 2000 cloud traffic records shows that the proposed EnsembleNet classifier obtains high performance, with an accuracy of 97.31%, precision of 96.90%, recall of 96.77%, F1-score of 96.83%, and Matthews Correlation Coefficient (MCC) of 0.943. The system effectively detects and blocks DDoS attacks while maintaining quality of service for legitimate traffic. The suggested EnhancedCloudDDoS-Guard framework has significant potential for smart DDoS detection and reduction in cloud settings. By integrating advanced feature selection and ensemble classification, the system improves detection accuracy and timeliness.