Résumé:
The rapid development of wireless networks with the massive increase in the number of
users and demands for high-speed data have pose real challenges in managing high frequency
bands such as millimeter waves, which offer high data rates, but their limited coverage is a
drawback which calls for the need for advanced technologies such as machine learning (ML).
In this study, we explore the use of various ML models to predict throughput performance
in 6G networks. We used the Lumos5G dataset, a widely recognized real-world dataset
after rigorous processing. We trained and tested these models, including Random Forest,
XGBoost, GradientBoosting, XTREE and advanced stacking with multiple meta-models,
optimizing their performance through the Optuna hyperparameter tuning framework. Fur
thermore, we generated synthetic samples using generative adversarial networks (GANs) to
address sample shortages and select key features to improve model‘s accuracy. The results
indicated significant improvements in throughput prediction accuracy, which was quantified
using metrics such as the root mean square error (RMSE) and coefficient of determination
(R2), confirming the effectiveness of the proposed technique
