QoS-Aware cloud security using lightweight EfficientNet with Adaptive Sparse Bayesian Optimization

Cloud security is critical for safeguarding data and services in cloud environments. Traditional deep learning methods focus primarily on improving attack detection accuracy but often neglect Quality of Service (QoS) parameters. These parameters, such as latency, bandwidth, and response time, are essential for the overall performance of cloud services. This research addresses the problem by proposing two novel methods: a lightweight EfficientNet deep learning model for accurate attack detection with minimal QoS impact and Adaptive Sparse Bayesian Optimization (ASBO) to improve hyperparameter tuning efficiency. ASBO reduces computational complexity by using sparse surrogate models, adaptive sampling, and early stopping mechanisms, ensuring the optimization process is efficient and suitable for real-time applications. The research objectives include enhancing attack prediction accuracy and QoS maintenance in cloud security. The study evaluates the proposed methods on the CICIDS2017, CICIDS2018, and UNSW-NB15 datasets, covering various attack types such as DDoS, Brute Force, SQL Injection, Botnet, Port Scanning, and Infiltration. The results demonstrate significant improvements over existing methods, achieving 5–7% higher accuracy in attack detection. The proposed EfficientNet + ASBO method also ensures better QoS, reduces latency, increases bandwidth efficiency, and improves response times compared to other models.