Answers are grounded in this paper and cite their sources.
This paper proposes an intrusion detection system (IDS) framework for imbalanced network traffic data, integrating supervised feature selection using Extra Trees (Extremely Randomised Trees), minority-class data synthesis via WGAN-GP (Wasserstein Generative Adversarial Network with Gradient Penalty), and sequential representation learning with a Transformer to enhance the detection of rare attack classes. First, Extra Trees is employed for dimensionality reduction and noise elimination, enabling the model to focus on attributes that are most informative for distinguishing malicious behaviours. Next, WGAN-GP generates high-quality synthetic samples for the minority class, improving the balance of the training distribution and mitigating bias toward the majority class. On the refined and balanced dataset, a Transformer encoder with a self-attention mechanism learns long-range dependencies within traffic sequences, thereby improving classification accuracy. Experiments on NF-CSE-CIC-IDS2018-v2 and FlowFeature show that the proposed method significantly improves the Recall and F1-score of attack classes, while also increasing overall efficiency in the classification stage through selective dimensionality reduction and enhanced input data quality. The results highlight the effectiveness of a sequential processing pipeline which is well-suited for deploying IDS in real-world data settings.
No takes yet. Share an insight, caveat, or question.
Danh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: