Deep learning for network security: an Attention-CNN-LSTM model for accurate intrusion detection.

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Network Security

Background:

• Evolving cyber threats necessitate advanced Intrusion Detection Systems (IDS).
• Traditional IDS struggle to keep pace with sophisticated attacks like malware and botnets.
• Deep learning offers potential for more effective network threat detection.

Purpose of the Study:

• To propose a novel hybrid deep learning model for enhanced network intrusion detection.
• To improve the accuracy and efficiency of Intrusion Detection Systems.
• To address the limitations of existing methods in identifying diverse cyber threats.

Main Methods:

• Developed a hybrid deep learning model named Attention-CNN-LSTM.
• Integrated Convolutional Neural Networks (CNNs) for spatial feature extraction.
• Utilized Long Short-Term Memory (LSTM) for temporal sequence modeling and a self-attention mechanism to prioritize informative features.

Main Results:

• Achieved high accuracy rates between 94.8% and 97.5% on NSL-KDD and Bot-IoT datasets.
• Significantly improved Matthews Correlation Coefficient (MCC) and F1-score.
• Demonstrated sub-35ms latency, supporting real-time inference and processing over 1200 records per second.

Conclusions:

• The Attention-CNN-LSTM model offers superior performance for Intrusion Detection Systems.
• The hybrid architecture effectively combines CNN, LSTM, and attention mechanisms for robust threat detection.
• The model's real-time capabilities make it suitable for high-traffic network environments.