Enhancing cybersecurity resilience through advanced machine learning techniques: a comprehensive analysis of threat detection and mitigation in critical infrastructures

Abstract

Cybersecurity is an increasingly critical concern due to the growing complexity and frequency of cyberattacks, particularly against critical infrastructures. Conventional security techniques such as intrusion detection and anomaly detection are often insufficient in addressing sophisticated threats like advanced persistent threats. This paper introduces a novel hybrid approach that combines LogBERT Transform for feature extraction, support vector machine for classification, and the Archimedes optimization algorithm for parameter tuning. This unique integration overcomes the limitations of traditional methods, improving detection accuracy while reducing false alarms. The proposed methodology is shown to outperform conventional systems, achieving 98.36% classification accuracy, 97.77% precision, 96.45% recall, and 97.10% F1-score. The model's low false positive and false negative rates demonstrate its practical feasibility in cybersecurity applications. This approach also enhances model interpretability and scalability, making it well-suited for deployment in large-scale networks and IoT systems, with further improvements in scalability and interpretability being key future directions.