Scholarlink

<p>This paper aims to introduce a performance comparison of multiple machine learning models for the prediction of throughput values in 5G networks. This research is of great importance in optimizing service quality and reducing congestion in cells. Predicting the throughput reveals the key parameters to maintain network resources and boost the user quality of experience. Furthermore, the research provides insights concerning activity performance in dimensionality reduction using principal component analysis (PCA). Moreover, the training dataset captures the network parameters’ patterns, environmental factors, and historical throughputs. The following algorithms were used to compare and evaluate the regression model: linear regression, random forest, LGBM, XGBoost, CatBoost, KNN, CART, and SVR. Finally, for each model, hyperparameter optimization was conducted using GridSearchCV, looking for the best hyperparameters. Model performance was qualified using RMSE, MAE, and R-score metrics on the test set. Metrics of prediction success and execution time were analyzed separately for the case of PCA and without PCA, for the case of hyperparameter optimization or without hyperparameter optimization. Inferences are valuable as they can be used to enhance network optimization and quality of service improvement strategies. The implemented PCA reduces computational complexity and training time, though it results in a slight decrease in prediction accuracy compared to models that do not utilize PCA. In conclusion, it is clear that in studies involving PCA, the problem of dataset complexity is effectively solved by reducing it to only a few principal components that best explain the variance in the dataset. Although there is a slight decrease in the predictive power of the model, this decrease is offset by improved computational performance factors, so that the approaches modified by PCA can be applied in real-time 5G fields.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Comparative Study of Machine Learning Regression Models with and Without Dimensionality Reduction for Predicting Throughput in 5G Networks

  • 期刊论文

  • 发表日期：

    2025年7月1日

查看全文
  • Mert Yağcıoğlu

摘要

This paper aims to introduce a performance comparison of multiple machine learning models for the prediction of throughput values in 5G networks. This research is of great importance in optimizing service quality and reducing congestion in cells. Predicting the throughput reveals the key parameters to maintain network resources and boost the user quality of experience. Furthermore, the research provides insights concerning activity performance in dimensionality reduction using principal component analysis (PCA). Moreover, the training dataset captures the network parameters’ patterns, environmental factors, and historical throughputs. The following algorithms were used to compare and evaluate the regression model: linear regression, random forest, LGBM, XGBoost, CatBoost, KNN, CART, and SVR. Finally, for each model, hyperparameter optimization was conducted using GridSearchCV, looking for the best hyperparameters. Model performance was qualified using RMSE, MAE, and R-score metrics on the test set. Metrics of prediction success and execution time were analyzed separately for the case of PCA and without PCA, for the case of hyperparameter optimization or without hyperparameter optimization. Inferences are valuable as they can be used to enhance network optimization and quality of service improvement strategies. The implemented PCA reduces computational complexity and training time, though it results in a slight decrease in prediction accuracy compared to models that do not utilize PCA. In conclusion, it is clear that in studies involving PCA, the problem of dataset complexity is effectively solved by reducing it to only a few principal components that best explain the variance in the dataset. Although there is a slight decrease in the predictive power of the model, this decrease is offset by improved computational performance factors, so that the approaches modified by PCA can be applied in real-time 5G fields.