As computer network security threats continue to grow in complexity, the need for more advanced security systems is obvious. Indeed, traditional methods of intrusion detection have struggled to keep pace with the changes and so researchers are looking to explore alternatives. A study in the International Journal of Computational Systems Engineering suggests that the integration of data augmentation and ensemble learning methods could be used to improve the accuracy of intrusion detection systems.
Xiaoli Zhou of the School of Information Engineering at Sichuan Top IT Vocational Institute in Chengdu, China, has focused on a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP). This is an advanced version of the standard machine learning model and can create realistic data through a process of competition between two neural networks. Conventional GANs often suffer from unstable training and pattern collapse, where the model fails to generate diverse data. The WGAN-GP variant mitigates these issues by incorporating a gradient penalty, according to the research, this helps to stabilize the training process and improve the quality of the generated data. It can then be used effectively to simulate network traffic for intrusion detection with a view to blocking hacking attempts.
There is the potential to enhance the WGAN-GP data quality still further by combining it with a stacking learning module. Stacking is an ensemble learning technique that involves training multiple models and then combining their outputs using a meta-classifier. In Zhou’s work, the stacking module integrates the predictions from several WGAN-GP models to allow them to be classified as normal or intrusive.
The approach was tested against well-established data augmentation methods, including the Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and a simple version of WGAN. The results showed that the WGAN-GP-based model had an accuracy rate of almost 90%, which is better than the scores for the other techniques tested. The model can thus distinguish between legitimate and potentially harmful network activity effectively. Optimisation might improve the accuracy and allow the system to be used to protect governments, corporations, individual, and others at risk from network security threats.
Zhou, X. (2024) ‘Research on network intrusion detection model that integrates WGAN-GP algorithm and stacking learning module’, Int. J. Computational Systems Engineering, Vol. 8, No. 6, pp.1–10.