Title: Ai-Driven Cybersecurity: Machine Learning Approaches For Secure Cloud And Network Systems
Authors: Venkata Surya Teja Gollapalli, Rajababu Budda, Rahul Jadon, Guman Singh Chauhan, Kannan Srinivasan, Akilan Selvaraj Saroja
Volume: 10
Issue: 2
Pages: 7-20
Publication Date: 2026/02/28
Abstract:
This paper proposes an AI-driven intrusion detection framework to enhance the security of cloud and network environments by integrating a hybrid Gated Recurrent Unit-Neural Turing Machine (GRU-NTM) model with pessimistic bilevel optimization. The proposed system addresses critical challenges such as class imbalance, temporal dependency modeling, and robustness against adversarial intrusions using the CIC-IDS-2017 benchmark dataset. A unified preprocessing pipeline incorporating data cleaning, min-max normalization, and SMOTE-based oversampling ensures high-quality and balanced input data. Mutual information-based feature selection and statistical flow aggregation are employed to improve discriminative learning. Experimental results demonstrate superior detection performance, achieving an accuracy of 99.87%, precision of 99.88%, recall of 99.85%, and an F1-score of 99.87%. The confusion matrix confirms highly reliable classification, with 25,576 benign and 25,567 DDoS instances correctly identified and fewer than 70 misclassifications. The proposed model attains an average precision of 0.99996 in the precision-recall analysis and an AUC of 0.99995 in ROC evaluation, indicating near-perfect detection capability. Additionally, system-level evaluation shows practical feasibility with a latency of 0.08 seconds, throughput of 198 requests per second, and CPU utilization of 39.9%, making the framework suitable for real-time deployment in dynamic cloud environments.