Navigating the Cloud: Unraveling Anomalies Using Self-Attention Based Conditional Generative Adversarial Network Approach in Data Center Networks

Anomaly detection in cloud data center networks is done through sophisticated machine learning and deep learning algorithms to detect performance, security, and resource utilization anomalies. The limitation is the high computational expense for real-time anomaly detection and analysis. In this paper, a self-attention-based conditional generative adversarial network (SA-CGAN) is proposed to detect anomalies in cloud data center networks. Initially, the input dataset is obtained from a publicly available dataset, such as the NSL-KDD dataset and the CDCAD dataset. After collecting the dataset, an Anisotropic Diffusion Kuwahara Filtering (ADKF)-based pre-processing approach is used to enhance the data quality. Additionally, to select the optimal features, an Enhanced FOX Optimization Algorithm (EFOXOA) is proposed, in which the FOX Optimization Algorithm is improved by using the Chimp Optimization Algorithm (COA). Moreover, the SA-CGAN is used to identify and categorize the abnormalities in the cloud network. The implementation of the suggested approach is carried out in Python, and the effectiveness of the presented approach is computed using various performance metrics. The proposed SA-CGAN-EFOXOA framework achieved superior anomaly detection performance on both the NSL-KDD and CDCAD datasets. On the NSL-KDD dataset, the proposed method attained classification accuracies of 98.91% for DoS (Denial of Service), 98.77% for Probe, 98.97% for R2L, 98.93% for U2R, and 98.65% for Normal traffic categories. Similarly, on the CDCAD dataset, the proposed framework achieved 98.25% accuracy for Bot attacks, 98.00% for DoS, 97.25% for DDoS, and 97.75% for Infiltration attacks. These results demonstrate the effectiveness of the proposed framework in accurately identifying diverse cloud network anomalies across multiple benchmark datasets.