Machine Learning Assisted Optimization of 5G Network Parametrization
Dominik Krah, Johannes Greul, Stefan Farthofer-Oster (2025): Machine Learning Assisted Optimization of 5G Network Parametrization In: 6th Interdisciplinary Data Science Conference (iDSC'25) Salzburg
In this work, we present the use of classical Machine Learning algorithms during our investigation of data rates achievable by individual clients in a mobile network with multiple communicating clients. In our study, we investigated the effect of differently parametrized 5G networks. In particular, we were interested in the effect of a custom prioritization parametrization on the overall performance of a network. Information on the utilization of a 5G cell is typically only available to the network provider and in very limited resolution, but was a central feature of our analyses. To obtain this information, we passively observed the network by an off-the-shelf software defined radio and estimated the cell utilization by a methodology based on Gaussian Mixture Modeling and k-means. Appropriate pre-processing and usage of lightweight, classical ML-models allowed us to run the estimation software on a Mini-PC, resulting in a versatile, mobile and cost-efficient measurement setup.
