AE-CIAM: a hybrid AI-enabled framework for low-rate DDoS attack detection in cloud computing

AE-CIAM: a hybrid AI-enabled framework for low-rate DDoS attack detection in cloud computing

연구 분야: Networking

학회: Cluster Computing

Cloud computing has gained popularity due to its scalability, cost-effectiveness, on-demand provisioning, pay-as-you-go billing, and enhanced accessibility. Recognizing these benefits, government agencies and industries are increasingly migrating their databases to the cloud. However, despite these strengths, cloud computing faces significant security challenges, including phishing, vulnerability exploitation, Distributed Denial-Of-Service (DDoS) attacks, and unauthorized access. Among these threats, DDoS attacks pose a particularly severe threat, by flooding servers with traffic, disrupting critical cloud services and rendering them inaccessible to legitimate users. In response to evolving threats, researchers are transitioning from traditional signature-based detection methods to machine learning and deep learning for DDoS detection. While these methods are effective for high-rate DDoS attacks, low-rate DDoS poses a challenge due to its subtle integration with legitimate traffic, device heterogeneity, diverse request specifications, and computational costs. To address these challenges, we propose a new hybrid framework called ‘AE-CIAM’ for the effective detection and classification of low-rate DDoS attacks in cloud environments. Our framework incorporates an Autoencoder (AE) enhanced with attention module, which adeptly extracts relevant features from data, eliminating the need for manual intervention and reducing feature space dimensionality. Unlike deep autoencoders, streamlined autoencoders with attention strike a balance between complexity and computational efficiency, offering versatility across diverse tasks. Subsequently, we employ the Convolutional Neural Networks (CNN) Inception with Attention Mechanism model to categorize attacks into different types of low-rate DDoS attacks, achieving high performance at a low computational cost. The proposed model demonstrates remarkable efficacy, achieving an accuracy exceeding 99.99% for binary classification and 99.44% for multiclassification on the CICIDS2017 dataset. Furthermore, we compare our AE-CIAM model with existing literature, showcasing its superiority through various performance metrics.