Authors: Assistant Professor S.Venkateswara Rao, B. Sindhuja
Abstract: With its ultra-high data speeds, low latency, enormous device connectivity, and better service dependability, fifth-generation (5G) cellular networks have greatly improved wireless communication. These networks have been rapidly deployed. But, the attack surface has grown with the ever-increasing complexity of 5G infrastructures and the ever-increasing volume of network traffic, leaving these networks open to cyber attacks, illegal access, and strange communication patterns. The ever-changing nature of 5G settings makes it difficult for traditional security measures to detect complex assaults as they happen. In light of these difficulties, this research introduces a framework for anomaly identification in 5G cellular networks that is based on machine learning. Various machine learning techniques are used by the suggested system. These techniques include Autoencoders, Random Forest, One-Class Support Vector Machine (One-Class SVM), and two ensemble learning methods. Both ensemble models include AdaBoost, Decision Tree, and Gradient Boosting into a Voting Classifier; however, the second model enhances anomaly detection performance by integrating AdaBoost, Decision Tree, and ExtraTree. When creating the models, we used a Cellular Dataset that included both typical and unusual 5G network data. Data cleansing, normalisation, and label encoding are some of the preprocessing activities that the dataset goes through before training in order to make the data better and the model work better. The suggested ensemble strategies are shown to outperform individual machine learning models in extensive experimental assessment. In complicated 5G communication scenarios, Ensemble 2 outperforms all other techniques with 100% Accuracy, Precision, Recall, and F1-score, proving its usefulness for real-time anomaly identification. To improve network security, reduce cyber threats, and enable intelligent monitoring of next-gen cellular communication systems, this study proposes a framework that is both scalable and dependable, all while maintaining the original machine learning methodology.
