Proposing a Cross‐Level Deep Neural Denial of Service/Distributed Denial of Service Attacks Detection in 5G Network Slicing

The 5G network is a new communication network development that mainly involves network slicing (NS) into isolated virtual networks dedicated to applications, industries, or user groups. Despite the flexibility provided by NS, it faces security challenges, especially denial‐of‐service (DoS)/distributed DoS attacks among different network slices on the 5G network. The proposed methodology in this study involves two levels of detection: at the slice level and at the cross level. Joint entropy with a dynamic threshold is used for early detection at the slice level, and a developed deep neural network (DDNN) model that uses a new activation function called the sigmoid–exponential–threshold activation function (SETAF) is employed for deep detection at the cross level. The CICIDS2017 dataset and NSDS are employed to build and evaluate the DDNN model. Results from the NS testbed show that the model effectively detects various attack patterns, achieving a detection accuracy of 99% with an average detection latency of approximately 49 ms. In addition, the results showed that using the new activation function led to a significant increase in accuracy, with most of the accuracy criteria reaching 0.994, and it also contributed to reducing training time to 33.1226 s and reducing mean square error to 0.0037. Thus, the SETAF function contributes to improving the detection of attacks in a dynamic manner that adapts to changes in network traffic.