DeepTransIDS: Transformer-Based Deep learning Model for Detecting DDoS Attacks on 5G NIDD

• A deep learning approach is proposed to classify attack types in 5G Networks, emphasizing on Transformer model-based approach. • CNN, RNN, and Transformer model-based approaches are employed to classify attacks in binary class classification scenarios for attack detection. Whereas, multi-class classification is for the identification of attacks. • A dataset generated from a real 5g-network including 1,215,890 records consisting of various types of attacks is discussed • The proposed work shows suitable applicability and accuracy for multi-class classification scenarios. With the rapid deployment of 5G technology, the security of advanced 5G networks is increasingly challenging. Conventional Intrusion Detection Systems (IDS) rely primarily on pre-5G datasets like NSL-KDD and CIC-IDS2017. These datasets cannot address the peculiar challenges that 5G brings, including low latency, large device density, and network slicing. The proposed DeepTransIDS implements a Transformer-based Intrusion Detection System to analyse network traffic in 5G non-IP data delivery scenarios. Unlike traditional IDS approaches that rely on Convolutional Neural Networks, this work uses the self-attention mechanism of Transformers to enhance the classification performance for multi-class network intrusion detection. The proposed model is trained on 5G-NIDD dataset with 1,215,890 network flows that include benign and a different type of malicious traffic. The transformer model achieves 99.79% multi-classification accuracy with better precision, recall, and F1-score, with an increase in accuracy of 0.10% compared to CNN-based IDS models, though the accuracy of RNN-based IDS model is 99.91% the computational time is significantly high. The confusion matrix analysis also confirms the model's ability to accurately identify intricate attack patterns even in the class of imbalance conditions. The findings confirm the dominance of the Transformer model in real-time intrusion detection in dynamic 5G networks.