AI-Enhanced Cyber Resilience Framework for Critical Infrastructure Protection Through Hybrid Deep Learning, Dynamic Risk Assessment, and Automated Recovery

Abstract

Critical infrastructure environments face increasingly sophisticated cy- ber threats capable of disrupting essential services, degrading opera- tional performance, and causing substantial economic losses. While re- cent advances in artificial intelligence have improved network attack identification, many existing approaches remain limited to classification tasks and offer insufficient support for impact evaluation, service restora- tion, and continuity management. To overcome these challenges, this study presents an AI-Enhanced Cyber Resilience Framework that com- bines adaptive feature engineering, hybrid deep learning, intelligent risk quantification, and recovery orchestration within a unified architecture. Initially, a Mutual Information-based strategy is employed to identify the most relevant network traffic attributes from heterogeneous cyber- security data. Subsequently, a CNN-BiLSTM-Attention architecture is developed to learn both local communication patterns and long-range behavioural dependencies associated with malicious activities. To sup- port resilience-oriented decision making, a Dynamic Risk Assessment Module incorporating the Traffic Intensity Score (TIS), Network Load Indicator (NLI), and Attack Pressure Score (APS) is introduced to eval- uate threat severity and infrastructure stress levels. Based on the gen- erated risk intelligence, an Automated Recovery Optimization Module prioritizes response actions and accelerates service restoration. Valida- tion is conducted using the UNSW-NB15 and CICIDS2017 benchmark datasets. Experimental findings demonstrate strong predictive capabil- ity, achieving 98.21% accuracy, 97.85% precision, 97.10% recall, 97.47% F1-score, and 0.992 AUC. Moreover, the proposed approach decreases the Mean Time to Recovery (MTTR) to 28.7 minutes, attains a Re- covery Success Rate (RSR) of 96.3%, preserves 98.1% system availabil- ity, and achieves a Resilience Index of 0.96. The results confirm that integrating intelligent analytics, adaptive assessment, and automated restoration significantly strengthens the robustness and sustainability of critical infrastructure operations under evolving cyber threats.