Non-Payload Based 5G Attack Detection with Qualitative Hybrid Feature Engineering

With the growing adoption of 5G technologies, the need for its cybersecurity while satisfying the latency requirements, is critical for end users. Feature selection is a popular method to reduce the amount of data required to be processed for traffic analysis and attack detection during the runtime. In this paper, we propose a two-stage 5G traffic analysis based attack detection framework. In the first stage, we utilize a hybrid feature selection approach to identify key features in 5G dataset such that the traffic analysis can be efficiently performed using only a handful of key features from the dataset. Hybrid feature selection incorporates a combination of filter and wrapper-based feature selection. Further, using these selected features, we designed a convolutional neural network based solution for attack detection for both binary and multiclass classification approach. We evaluated our feature selection and attack detection approach on 5G-NIDD benchmark dataset, and achieved state-of-the-art accuracy with around 99% and 91% accuracy for non-strict and strict qualitative feature set, respectively.