A hybrid neuro-fusion model for network intrusion detection using non-symmetric deep autoencoders and transformer-augmented gradient boosting

NIDS face major challenges in the ever-changing cyber landscape, primarily because of the spreading and increasing rate of network threats. Feature redundancy, scalability issues and the dynamism of attack patterns are common problems that arise when dealing with conventional machine learning models. Hence, this paper proposes a Hybrid Neuro Fusion Model that can overcome these limitations. The model combines Deep Feature Extraction with Transformer Augmented Gradient Boosting (TAGB) to obtain a strong and interpretable classifier. In order to further improve gradient boosting performance, the contextual attention over the feature space is provided through a transformer module, while the latent representations of complex network traffic are captured through an NSDAE component. The CICIDS2018 dataset, which is currently recognized as a popular benchmark for intrusion detection, was used to test the proposed architecture. Experimental results show that the hybrid model performs better than more traditional models (like deep learning baselines), Random Forests and Support Vector Machines. The model achieves a good detection accuracy of 99.25%, precision of 98.91%, recall of 99.17%, and F1-score of 99.02% with a significant decrease of false positives at the same time. These results highlight the model&s;s ability to identify elaborate and advanced attack patterns and thus indicates its potential for use in online intelligent NIDS deployments.