Payload-based 5G Attack Detection

Due to the versatility and flexibility of 5G technology, its global adoption is increasing rapidly. The enhanced configurability of 5G-Core network makes it susceptible to attacks due to potential errors in complex configurations. As the web-based technologies and protocols used in the 5G Core architecture are widely adopted, additional efforts are essential to maintain the security of the 5G system. Therefore, 5G attack detection has received keen interest from research community. Machine learning based solutions provide an effective way of determining a potential threat and can function as a part of a versatile defense. In this paper, we proposed an ensemble approach for attack detection using Random Forest, Logistic Regression and Convolutional Neural Networks and performed evaluation using the 5GAD dataset. We evaluated the performance of each individual model component and compared them against the final ensemble approach. We further investigated the benefit of embedding layer on our proposed CNN architecture within the ensemble. We evaluated the performance of our methodology and achieved around 99% accuracy for the proposed method.