Navigating Security Threats in Cloud-Based Systems: A Hybrid BiLSTM-GRU Intrusion Detection Framework

Abstract

Cloud infrastructures face exponentially increasing cybersecurity risks with 75% surge in cloud intrusions annually and 4 out of 5 security vulnerabilities originating from cloud environments [1]. This paper addresses the critical challenge of real-time intrusion detection in multitenant cloud environments by proposing a novel lightweight BiLSTM-GRU hybrid neural architecture. Unlike existing approaches that sacrifice either accuracy or inference speed, this method achieves 96.7% accuracy (±2.1% at 95% CI) with 95.1% F1-score, 12 ms inference latency, and an exceptionally low 0.03% false-positive rate. The hybrid model outperforms CNN-LSTM baselines by 3.6% accuracy and reduces latency by 36% while maintaining 0.824 Matthews Correlation Coefficient [3]. Comprehensive evaluation on ISOT Cloud IDS and CIC-IDS 2018 datasets with statistical significance testing (p less than 0.001) validates production-ready performance on commodity GPU hardware. Key contributions include: (1) a mathematically rigorous BiLSTM-GRU fusion architecture optimized for cloud traffic patterns, (2) comprehensive ablation studies demonstrating component-wise performance gains, (3) real-time GPU deployment validation with resource utilization analysis, and (4) statistical robustness verification through 10- fold cross-validation with confidence intervals. These findings establish a new state-of-the-art for cloud-native intrusion detection systems with practical deployment feasibility for enterprise environments.