TL;DR:
- RUDN University mathematicians compared SARIMA and Holt-Winter models for traffic forecasting in 5G and 6G networks.
- Both models proved effective but excelled in different scenarios.
- SARIMA was more accurate for user-to-base station traffic, while Holt-Winter was better for base station-to-user traffic.
- Future research will combine statistical models and machine learning for enhanced predictions.
Main AI News:
In the fast-evolving sphere of 5G and the impending 6G networks, the imperative lies in managing traffic load and optimizing resource allocation in real time. The prerequisite for these networks is the ability to continually monitor current metrics and foresee potential developments. This capability is pivotal for services to make informed decisions regarding network segmentation and load distribution. Machine learning models have traditionally served as the linchpin in this predictive process.
RUDN University’s mathematicians have undertaken a meticulous comparison of two forecasting models, meticulously dissecting their strengths and weaknesses. Their insightful research findings have been disseminated in the esteemed pages of the Future Internet journal.
“5G and the forthcoming 6G networks will serve as the infrastructure for a diverse array of applications, including drones, virtual and augmented reality. However, the surge in the number of connected devices leads to a dramatic spike in network traffic, resulting in network congestion. Consequently, this hampers service quality, leading to network latency and data loss. Therefore, network architecture must dynamically adapt to varying traffic volumes while considering different traffic types with distinct requirements,” emphasized Dr. Irina Kochetkova, a distinguished Associate Professor at the RUDN Institute of Computer Science and Telecommunications.
The crux of the mathematicians’ study revolved around the comparison of two time series analysis models—the Seasonal Integrated Autoregressive Moving Average (SARIMA) model and the Holt-Winter model. The cornerstone of their analysis was a comprehensive dataset sourced from a Portuguese mobile operator, encompassing traffic volumes for both download and upload operations over fixed one-hour intervals.
Both models demonstrated commendable prowess in forecasting traffic for the ensuing hour. Nevertheless, SARIMA stood out as the preferred choice for predicting traffic from the user to the base station, boasting an average error rate of 11.2%, a noteworthy 4% lower than the alternative model. In contrast, the Holt-Winter model excelled in predicting traffic from the base station to the user, exhibiting an error rate of 4.17%, significantly outperforming the second model’s 9.9% error rate.
“In the realm of traffic forecasting, there exists no one-size-fits-all solution. Both SARIMA and the Holt-Winter model excel in their respective domains, catering to distinct traffic prediction scenarios. Each dataset warrants a tailored approach. Our future research trajectory will entail the integration of statistical models with machine learning techniques, promising more precise forecasts and enhanced anomaly detection capabilities,” elucidated Associate Professor Kochetkova.
Conclusion:
The study highlights the importance of tailored forecasting models for different aspects of 5G and 6G network traffic. This information is crucial for businesses operating in the telecommunications industry, as it emphasizes the need for adaptable network architectures to accommodate diverse traffic types. Companies should invest in predictive capabilities to ensure efficient resource allocation and high-quality service delivery in the evolving landscape of mobile networks.
