A new deep-learning architecture could deflect cyberattacks by combining several methods for analysing network traffic, according to research in the International Journal of Business Intelligence and Data Mining. The approach addresses a weakness in standard intrusion-detection systems, which struggle with the volume and complexity of modern network traffic.
The researchers combined a deep neural network (DNN), which identifies complex patterns in large sets of features, with a bidirectional gated recurrent unit (BiGRU), a type of neural network designed to identify relationships across sequences in both directions. Both components use an attention mechanism, which allows the system to concentrate on the features and points in a sequence that are most relevant to anomaly detection. The outputs are brought together and classified using a multilayer perceptron (MLP), another neural-network architecture used to assign data to categories.
Tests on two well-known benchmark datasets showed that the system could achieve validation accuracies of around 99 per cent. The researchers explain that the system is stable in training and has comparatively strong robustness and generalisation, which means it should be effective on data beyond its training data.
The results suggest that combining static-feature analysis with temporal pattern recognition can harden cybersecurity systems against denial-of-service attacks, network probing, and attempts to gain unauthorised access.
Yan, H., Liu, H., Yu, P., Xu, X., Li, M., Long, Y., Chen, H., Wang, Q. and Long, D. (2026) ‘DNN and BiGRU-based hierarchical attention network for intrusion detection’, Int. J. Business Intelligence and Data Mining, Vol. 28, No. 10, pp.1–24.
News media may use this press release as source material, in whole or in part, provided the content is not materially misrepresented. A link back to the original article is appreciated.
No comments:
Post a Comment