نور العابدي

Unmatched security issues brought on by service-based architectures, network slicing, an‎d large device connectivity are introduced with the deployment of fifth-generation (5G) mobile networks. This study proposes a feature-optimized machine learning-based intrusion detection system designed for 5G networks an‎d keeps remarkable detection performance. Using a strict "common characteristics" approach, the research removes environment-specific identifiers an‎d data leakage sources that could threaten model generalizability while also systematically identifying 13 generalizable behavioral features from network flow data. Seven machine learning algorithms XGBoost, Ran‎dom Forest, Decision Tree, Artificial Neural Networks, K-Nearest Neighbors, Logistic Regression, an‎d Naive Bayes are thoroughly examined on both the 5G-NIDD dataset an‎d CICDDoS2019 dataset for binary an‎d multi-class classification assignments. Experimental results show that XGBoost maintains computing efficiency with training times of 4.484 seconds an‎d prediction times of 0.116 seconds while achieving excellent performance with 99.97% accuracy on 5G-NIDD binary classification an‎d 99.96% accuracy on multi-class classification. Achieving somewhat quicker inference time of 0.040 seconds an‎d equivalent accuracy of 99.96%, decision tree presents as the best option for resource-constrained deployments. Achieving 99.96% accuracy on CICDDoS2019 binary classification, the feature-optimized approach successfully generalizes across datasets, hence confirming the robustness of chosen features beyond 5G-specific contexts. This study uses a structured feature categorization system that groups 13 important features into four unique categories: Temporal, Volume, Statistical,an‎d Rate. This is done to improve detection accuracy an‎d generalizability. This technique methodically identifies behavioral characteristics, such as timing abnormalities an‎d volumetric irregularities, such that different attack vectors can be found while removing environment-specific identifiers to make the model more robust.

امنيت شبكه 5G , سيستم‌هاي تشخيص نفوذ , بهينه‌سازي ويژگي , يادگيري ماشين , طبقه‌بندي ترافيك شبكه

5G Network Security , Intrusion detection systems , Feature Optimization , Machine Learning , network traffic classification

Noor Abedi

Dr. Akbari