CiNeT: A Comparative Study of a CNN-Based Intrusion Detection System with TensorFlow and PyTorch for 5G and Beyond

As 5G and beyond networks grow in heterogeneity, complexity, and scale, traditional Intrusion Detection Systems (IDS) struggle to maintain accurate and precise detection. Deep Learning (DL) provides a promising alternative detection method however, DL based IDS suffer from issues relating to interpretability, performance variability tied to framework dependency and high computational overhead which limit and question their deployment in real-world applications. In this study we introduce a novel new DL based IDS employing Convolutional Neural Networks (CNN) to identify different types of attack together with a bijective encoding-decoding pipeline that acts between network traffic features (such as IPv6, IPv4, MAC addresses and protocol data), and their RGB representations that facilitates our DL IDS in detecting spatial patterns without sacrificing fidelity. Following the application of our bijective pipeline from RGB images to their corresponding network traffic features the 'black-box' problem is resolved and digital forensic traceability enabled. Finally, we have evaluated our DL IDS on three datasets, UNSW NB-15, InSDN, and ToN\textunderscore IoT, analysing the accuracy, GPU usage, memory utilisation, training, testing, and validation time. To summarise, in this study, we present a new CNN based IDS with an end-to-end pipeline between network traffic data and their RGB representation that offers high performance and transparency.