Dynamic Anomaly Detection in 5G-Connected IoT Devices using Transfer Learning

The need for effective anomaly detection techniques to safeguard interconnected systems has increased as a result of the Internet of Things' (IoT) rapid proliferation and the introduction of 5G networks. This study offers an innovative approach for building a flexible machine learning model for effective 5G-connected IoT anomaly detection via transfer learning. The study illustrates cross-domain knowledge transfer by leveraging the common known NUSW-NB15 and 5G-NNID datasets. The approach in this contribution covers a range of machine learning models, including both supervised and unsupervised learning methods, In order to use the most effective pretrained model in our Transfer Learning model. These models are evaluated with performance indicators like Fl, AUC-PR, and AUC-ROC. A Transfer Model that combines features from both datasets demonstrates adaptability and robustness and has been verified. By utilizing shared and similar features extracted from both datasets, this model seeks to generalize its operation across various 5G Wireless communication scenarios. The study offers technical recommendations for transfer model development and evaluation for improving anomaly detection in the 5G IoT environment. In this evaluation, we discovered that among the supervised learning models, Neural Networks (NN) and Convo-lutional Neural Networks (CNN) stood out as champions. They both achieved an impressive accuracy of nearly 99% in both the NUSW-NB15 and 5G-NNID datasets.