Network anomaly detection using Deep Autoencoder and parallel Artificial Bee Colony algorithm-trained neural network

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Machine Learning

Background:

• Cyberattacks are increasingly complex, challenging traditional intrusion detection systems.
• Machine learning and deep learning are active research areas for network intrusion detection.
• Conventional training algorithms for Artificial Neural Networks (ANNs) face optimization issues like local minima and slow convergence.

Purpose of the Study:

• To develop an improved algorithm for training Artificial Neural Networks (ANNs) for network intrusion detection.
• To address the limitations of conventional training methods and metaheuristics in ANNs.
• To enhance the detection rate (DR) and reduce the false alarm rate (FAR) in network security.

Main Methods:

• Implementation of a Deep Autoencoder (DAE) combined with a vectorized and parallelized Artificial Bee Colony (ABC) algorithm.
• Training feed-forward Artificial Neural Networks (ANNs) using the proposed DAE-based parallel ABC algorithm.
• Testing the proposed method on the UNSW-NB15 and NF-UNSW-NB15-v2 datasets for network intrusion detection.

Main Results:

• The DAE-based parallel ABC-ANN demonstrated superior performance compared to existing metaheuristics.
• On the UNSW-NB15 dataset, the proposed approach increased detection rate (DR) from 0.76 to 0.81 and reduced false alarm rate (FAR) from 0.016 to 0.005.
• On the NF-UNSW-NB15-v2 dataset, the FAR was reduced from 0.006 to 0.0003 compared to the ANN-BP algorithm.

Conclusions:

• The proposed Deep Autoencoder (DAE)-based parallel Artificial Bee Colony (ABC) algorithm effectively enhances network intrusion detection.
• The approach significantly improves detection accuracy and reduces false alarms, outperforming conventional methods.
• This work offers a promising solution for strengthening network security against sophisticated cyber threats.