Leveraging self attention driven gated recurrent unit with crocodile optimization algorithm for cyberattack detection...

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Cybersecurity is crucial for protecting data from diverse cyberattacks like malware and DoS attacks.
• Artificial Intelligence (AI) and Machine Learning (ML) offer novel techniques for enhancing cybersecurity defenses.
• Federated Learning (FL) addresses challenges in security, data privacy, and access rights within ML models.

Purpose of the Study:

• To propose an effective real-time cyberattack prevention methodology using Federated Learning and advanced optimization algorithms.
• To introduce the Self-Attention Mechanism-Driven Federated Learning for Secure Cyberattack Detection with Crocodile Optimization Algorithm (SAMFL-SCDCOA).

Main Methods:

• Data preprocessing using Z-score normalization for consistency and accuracy.
• Feature selection (FS) employing the Crocodile Optimization Algorithm (COA).
• Cybersecurity classification using a Gated Recurrent Unit with Self-Attention (GRU-SA) model, optimized by the Improved Pelican Optimization Algorithm (IPOA).

Main Results:

• The SAMFL-SCDCOA methodology demonstrated superior performance in cyberattack detection.
• Experimental validation on the CICIDS-2017 dataset confirmed the technique's effectiveness.
• Achieved a classification accuracy of 99.04%, outperforming existing models.

Conclusions:

• The proposed SAMFL-SCDCOA approach offers a robust solution for real-time cyberattack detection.
• The integration of FL, AI optimization, and advanced ML models significantly enhances cybersecurity.
• This methodology provides a promising direction for securing large-scale systems against evolving cyber threats.