Metaheuristic optimized deep learning for intelligent intrusion detection in cloud environments

Abstract

Designing intrusion detection systems for cloud environments requires a framework that not only achieves high accuracy but also effectively identifies a wide spectrum of attacks, including DoS/DDoS, Probe, R2L, and U2R, under dynamic and noisy network conditions. This paper presents the integrated AQSE-QDST framework, which combines the Adaptive Quantum Swarm Evolution (AQSE) algorithm for feature space optimization with QDST-Net (Quantum-Inspired Dual Spatial-Temporal Network) as the classification engine. In addition, the Entropy-Guided Adaptive Flow Normalization (EAFN) mechanism is incorporated to accelerate convergence and reduce fitness fluctuations during the early stages of training. This three-layer design enables robust feature extraction, dimensionality reduction, and stable feature selection for diverse and imbalanced datasets. Experiments conducted on three benchmark cybersecurity datasets demonstrate that the proposed framework performs effectively in detecting both frequent and rare attacks. The model achieves accuracy rates of 99.69% on NSL-KDD, 98.86% on CIC-IDS2017, and 98.65% on UNSW-NB15, highlighting its capability to detect both high-volume attacks such as DoS/DDoS and Probe and low-frequency attacks such as R2L and U2R. Furthermore, convergence analysis indicates that AQSE-QDST outperforms baseline methods by maintaining more stable fitness values and more consistent feature selection behavior.