PSO-Enhanced Federated CNN-LSTM for Privacy-Preserving DDoS Detection in Multi-Tenant Clouds

Abstract

Distributed Denial-of-Service (DDoS) attacks are a major security challenge in cloud computing because of their characteristic of depleting resources, thereby lowering availability of services. The centralized machine learning based detection of DDoS attack poses some major challenges related to privacy and scalability issues within multi-tenant cloud. This work introduces a PSO-Optimized Federated CNN-LSTM model for accurate and privacy-preserving DDoS detection. CNN used for extraction of spatial traffic features, LSTM is used for capturing sequential traffic behaviors. Federated learning permits collaborative training of the model without exposing raw traffic. Particle swarm optimization (PSO) is applied for hyper parameter optimization. Evaluation on DRDoS data-set consisting of 193,956 traffic records and 70 features provides a 98.62% detection accuracy. Our model provided the increase in scalability, privacy-preservation, and correct DDoS detection.