Ensemble of feature augmented convolutional neural network and deep autoencoder for efficient detection of network...

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Effective network traffic monitoring is crucial for detecting intrusions and cyberattacks.
• Existing Network Intrusion Detection Systems (NIDS) require enhanced packet flow classification efficiency.

Purpose of the Study:

• To propose a novel ensemble of deep learning techniques to improve packet flow classification in NIDS.
• To enhance the detection of minority attack classes within network security datasets.

Main Methods:

• Developed a three-phase approach: Feature Augmented Convolutional Neural Network (FA-CNN), Deep Autoencoder, and an ensemble of both.
• FA-CNN utilizes augmented features selected via Mutual Information.
• Ensembled FA-CNN with Deep Autoencoder for a robust classification model.

Main Results:

• Experimental validation on NSL-KDD and CICIDS2017 datasets demonstrated superior performance compared to existing methods.
• Achieved an overall accuracy of 97% on the NSL-KDD dataset and 95% on the CICIDS2017 dataset.
• Significantly improved the detection rate for minority attack classes, such as U2R (NSL-KDD) and Heartbleed (CICIDS2017).

Conclusions:

• The proposed deep learning ensemble method offers a significant advancement in NIDS packet flow classification.
• The approach effectively addresses the challenge of detecting low-frequency, high-impact network attacks.
• This work provides a more accurate and efficient solution for real-time network security monitoring.