Protection for 5G Network Access through Data-driven Deep Neural Network Clustering

Abstract

This study presents an innovative security model for wireless access in 5G networks, referred to as 5GDoSec. Considering that a concern within the security of 5G network access pertains to Distributed Denial of Service (DOS) attacks attributed to its orientation towards the Internet of Things (IoT), a security model is put forth. This novel model offers a resolution to this predicament, demanding minimal user data, user-friendly operation, streamlined training and configuration, as well as modest computational demands and exceptional adaptability. The primary target of this model is to identify potential intruders and malicious actors through the application of Deep Neural Networks coupled with machine learning methodologies. The methodology follows an evolutionary process based on prototypes where an unsupervised security model is built through data analysis. This approach leverages access data collected from a specific entry point that aggregates, profiles, and categorizes authorized network users. The aim is to discern, based on access metrics and active durations, those individuals that might pose a security risk. The adaptable nature of the 5GDoSec model has been empirically demonstrated and stands as a dependable means of accurately categorizing hazardous users. Empirical validation, gauged through the DaviesBouldin index, underscores its superiority over alternative techniques such as Kmeans and Linkage.