Slice-Specific Machine Learning Models for Intrusion Detection in 5G Telecommunication Networks

The security challenges introduced by 5G network slicing demand tailored intrusion detection systems (IDS). Traditional intrusion detection systems (IDS) and intrusion detection and prevention systems (IDPS) frameworks, built for static network configurations, are inadequate for the dynamic and heterogeneous nature of 5G networks. To address this gap, this study develops and evaluates slice-specific machine learning models to enhance intrusion detection across different 5G slices, namely: enhanced Mobile Broadband (eMBB), massive Machine-Type Communication (mMTC), and Ultra-Reliable Low-Latency Communication (URLLC). Random Forest, Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) models were applied to publicly available datasets representing each slice. These models are assessed based on their accuracy, precision, recall, F1-score, area under the receiver operating characteristic curve (AUC-ROC), confusion matrix and execution time. The results reveal that the LSTM model achieved the highest accuracy and AUC-ROC scores for the eMBB and mMTC slices, making it suitable for applications where detection accuracy is critical despite higher computational demands. In contrast, Random Forest demonstrated superior computational efficiency, making it the most preferred model for latency-sensitive URLLC slice, where real-time detection is essential. While the SVM model performed well in terms of accuracy, its high computational cost renders it less practical for real-time applications, particularly in URLLC environments. This research provides insights for enhancing 5G network security through the deployment of slice-specific machine learning models, thereby addressing the critical need for adaptable and efficient IDS frameworks.