Composition of Hybrid Deep Learning Model and Feature Optimization for Intrusion Detection System.

Area of Science:

• Cybersecurity
• Network Security
• Machine Learning for Network Defense

Background:

• The proliferation of Internet of Things (IoT) devices has expanded network attack surfaces, necessitating robust cybersecurity measures.
• Intrusion Detection Systems (IDSs) are crucial for identifying malicious network traffic, with zero-day attack detection being a significant challenge.
• Deep learning (DL) methods show promise for IDSs, but effective feature learning remains an open problem.

Purpose of the Study:

• To propose and evaluate a novel IDS technique that optimizes feature learning for improved malicious traffic identification.
• To investigate the efficacy of combining Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU) for enhanced IDS performance.
• To address the challenge of zero-day attack detection in increasingly complex network environments.

Main Methods:

• A hybrid deep learning model integrating CNN and GRU architectures was developed.
• Various CNN-GRU sequence combinations were explored to optimize network parameters for feature extraction.
• The proposed model was trained and validated using the CICIDS-2017 benchmark dataset.

Main Results:

• The proposed IDS achieved a high detection accuracy of 98.73% for various network attacks.
• The system demonstrated a low False Positive Rate (FPR) of 0.075.
• Comparative analysis showed significant improvements over existing IDS techniques.

Conclusions:

• The developed CNN-GRU based IDS offers a promising solution for enhancing network security and detecting sophisticated cyber threats.
• The proposed approach effectively addresses the feature learning challenges in deep learning-based IDSs.
• The findings indicate the practical efficacy of the proposed IDS for real-world cybersecurity applications.