AI-Driven Intrusion Detection Systems: Enhancing Cybersecurity with Machine Learning Algorithms

Abstract

Cybersecurity has become a critical issue due to the increasing sophistication and frequency of cyber-attacks. Traditional intrusion detection systems (IDS) have limitations, particularly in detecting new and unknown attacks. This paper proposes a hybrid approach combining machine learning (ML) and attention mechanisms in an AI-driven IDS to address these limitations. The proposed system utilizes autoencoders for anomaly detection and integrates attention mechanisms to focus on the most relevant features, improving detection accuracy and reducing false positives. Results indicate that the proposed system outperforms traditional methods, demonstrating a significant improvement in accuracy by 15-20%, alongside a reduction in false positive rates. The use of Z-score normalization in data preprocessing further enhances the system's ability to process and detect intrusions effectively. The proposed system shows promising results, offering an adaptive and scalable solution to evolving cybersecurity threats. Future work involves exploring the integration of more advanced models like transformers, improving system scalability, and addressing the challenge of model explainability for better transparency and trust in security operations.