Mathematicians compare machine learning models to predict 5G and 6G traffic

5G and 6G networks must account for load and constantly adapt resource consumption. To do this, you need to be able to track and predict current metrics. This is how the service decides to divide the network into slices and distribute the load. Machine learning models are typically used for prediction.

Mathematicians from RUDN University compared two predictive models and showed the advantages and disadvantages of each.Their research will be published in the journal future internet.

“5G and 6G networks will support drones, virtual reality, and augmented reality. Additionally, as the number of connected devices increases, traffic will surge and network congestion will occur. As a result, the quality of service will decrease. “The network architecture will therefore have to adapt to the traffic volume and take into account several types of traffic with different requirements.” RUDN Computer Science and Electrical said Dr. Irina Kochetkova, Associate Professor at the Institute of Communication.

Mathematicians compared two time series analysis models: the Seasonally Integrated Autoregressive Moving Average (SARIMA) model and the Holt-Winter model. To build the model, we used data from a Portuguese mobile operator on the amount of download and upload traffic over a period of time (1 hour).

Both models were found to be good at predicting traffic volume for the next hour. However, his SARIMA was better at predicting traffic from users to base stations, with an average error of 11.2%, a 4% reduction over the second model. The Holt-Winter model was better at predicting traffic from base stations to users, with an error of 4.17% instead of 9.9%.

“Both models are effective in predicting traffic averages, but the Holt-Winters model is better suited for predicting traffic from base stations to users, and SARIMA is better suited for predicting traffic from users to base stations. There is no single model. “Each dataset requires its approach, so we need a one-size-fits-all solution here. Future research will explore the use of statistical models to achieve more accurate predictions and anomaly detection.” and machine learning techniques,” said Associate Professor Kochetkova.