Enhancing Network Intrusion Detection for Big Datasets with Hybrid Deep Learning: A SWIN Transformer and VGG19

The rapid evolution of cyber threats has rendered classical methods for detecting cyber-attacks increasingly vulnerable and time-consuming. As cyber threats grow in complexity and scale, timely and accurate detection becomes critical, yet challenging, especially when dealing with massive and heterogeneous data. Existing systems have explored various machine-learning methods to analyze and predict threats, but classical techniques struggle to scale effectively in big data environments. In response, anomaly-based intrusion detection systems have garnered significant attention, particularly in the context of big data. This research introduces a novel approach that leverages the Adaptive Synthetic (ADASYN) sampling technique to generate new minority instances and a KNN-based under-sampling method to eliminate overlapping instances, ensuring a balanced and refined dataset. To further enhance detection accuracy, we propose a hybrid deep learning model that combines the SWIN Transformer and VGG19 architectures. This approach is specifically designed to handle the vast and complex datasets typical of modern cybersecurity challenges. The efficacy of the proposed method is evaluated using the UNSW _NB15 and NSL-KDD datasets, representing large-scale and diverse network traffic. Our approach achieves a remarkable accuracy of 99.84% on the NSL-KDD dataset and 88.34% on the UNSW _NB15 dataset. These results demonstrate the capability of the proposed model to effectively detect network intrusions in big data environments, outperforming existing methods and offering a robust solution to modern cybersecurity challenges.