A hybrid feature weighted attention based deep learning approach for an intrusion detection system using the random...

Area of Science:

• Cybersecurity
• Machine Learning
• Network Security

Background:

• Network systems face evolving threats, making intrusion detection challenging.
• Class imbalance in intrusion detection datasets leads to high false-negative rates.
• Existing Intrusion Detection Systems (IDS) require improved accuracy and reduced false alarms.

Purpose of the Study:

• To develop an advanced hybrid model for effective intrusion detection.
• To address the critical issue of class imbalance in network traffic data.
• To enhance the sensitivity of IDS to intrusions while minimizing false positives.

Main Methods:

• A hybrid model (MCL-FWA-BILSTM) integrating Mean Convolutional Layer (MCL), Feature-Weighted Attention (FWA), Bidirectional Long Short-Term Memory (BILSTM), and Random Forest.
• Preprocessing and feature extraction using CNN-MCL layers.
• Mitigation of class imbalance using BI-LSTM and self-attention feature weights.
• Classification using Random Forest on concatenated attention and BI-LSTM features.

Main Results:

• Achieved 99.67% (binary) and 99.88% (multi-class) accuracy on NSL-KDD dataset.
• Achieved 99.56% (binary) and 99.45% (multi-class) accuracy on UNSW-NB15 dataset.
• Outperformed previous models in detection rate, False Positive Rate (FPR), and F-score, reducing false positives and increasing true positives.

Conclusions:

• The MCL-FWA-BILSTM model effectively handles class imbalance in intrusion detection.
• The proposed model demonstrates superior performance in identifying diverse network intrusions.
• This research offers valuable insights and practical solutions for robust IDS development in cybersecurity.