Abstract: The Citizens Broadband Radio Service (CBRS) system requires an optimal network slicing approach to allocate resources and manage quality of service (QoS) for various applications. This paper focuses on refinement of network slicing through non-linear classification methods: Decision Trees, Random Forests, and K-Nearest Neighbours (KNN). We study how these algorithms meet accuracy and processing time benchmarks best suited for dynamic spectrum access situations. With the aid of machine learning, we build a model that predicts channel conditions and classifies network slices in real-time based on traffic flows. The results indicate that Random Forest achieved the highest accuracy but Decision Trees provided a favourable balance between computation and precision. KNN performed better than many of the other algorithms in terms of accuracy but cost more time in processing. The balance between best accuracy and lowest computational time highlights classification methods that could be implemented for real-time CBRS network slicing. Consequently, the proposals made enhance spectral efficiency while reducing latency and improving overall performance across the network.

Keywords: CBRS, Network Slicing, KNN, Decision Tree, Random Forest

Pages: 482-488