A Comparative Analysis of Feature Selection and Machine Learning Algorithms for Enhanced Anomaly Detection in 5G Core Networks

A critical challenge lies in choosing the appropriate processing techniques to achieve optimal evaluation performance in machine learning algorithms for the 5G core data analysis in anomaly detection use case. In this work, we address this challenge by focusing on the comparative analysis of Machine Learning (ML) algorithms combined with data processing techniques for anomaly detection in 5G core networks. ML algorithms are promising for detecting such anomalies, but their performance heavily depends on the quality of the features used. This work highlights the importance of feature selection, demonstrating that by applying techniques such as Pearson’s correlation and dimensionality reduction (PCA and LDA), we can enhance the detection accuracy of various ML models. We evaluate eight ML algorithms using the 5GAD-2022 cybersecurity dataset and show that Random Forest achieves the best performance in terms of F1-score, AUC, and accuracy, regardless of the preprocessing techniques applied. The results suggest that meticulous feature selection can significantly enhance the algorithms’ effectiveness in detecting anomalies.