Scholarlink

This research explores the application of predictive analytics through Machine Learning (ML) algorithms to enhance Mobile Network Key Performance Indicators (KPIs), specifically focusing on Reference Signal Received Power (RSRP) as coverage and Reference Signal Received Quality (RSRQ) as quality. Various regression and classification modelling techniques were applied to drive-test measurements collected around the University of Hull, utilizing supervised ML algorithms such as Decision Tree (DT), Logistic Regression (LogisticR), Random Forest (RF), Support Vector Machine/Regressor (SVM/SVR), Light Gradient Boosting Machine (LightGBM), K-Nearest Neighbour (KNN), Extra Trees (ET), Extreme Gradient Boosting (XGB), Multi-Layer Perceptron (MLP), Deep Neural Network (DNN), Gaussian Naïve Bayes (GNB), and Gradient Boosting (GB) to benchmark the performance of four Mobile Network Operators (MNOs)/Mobile Virtual Network Operators (MVNOs) at various locations around the University of Hull, with additional model validation conducted in Hull City Centre, Barton Upon Humber, and Newland as use cases. The Random Forest (RF) model emerged as the best-performing algorithm, achieving a Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) below 3.7, a Mean Absolute Percentage Error (MAPE) under 7.03, a Coefficient of Determination (R2) greater than 74%, a Receiver Operating Characteristic Area Under the Curve (ROC_AUC) above 93%, and an Accuracy exceeding 82%. Additionally, the ensemble learning (EL) model, which combined the strengths of RF, GB, ET, SVR, XGB, and LightGBM for regression, and LogisticR, SVM, MLP, GB, ET, and RF for classification, delivered an overall performance with RMSE and MAE below 4, R2 above 72%, accuracy exceeding 81%, and ROC_AUC above 85%. This highlights the EL model’s ability to predict network coverage (RSRP) and quality (RSRQ) as excellent, good, fair, bad, or poor with high precision. This study demonstrates the uniqueness of integrating multiple KPIs (RSRP and RSRQ) and prediction techniques (regression and classification) within an Artificial Intelligence (AI)-driven solution, providing a robust framework for improving network performance, particularly in scenarios where data collection through drive testing is limited.

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

Ensemble Supervised Learning-Based Approaches for Mobile Network Coverage and Quality Predictions in a University Setting

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Morrison Okiemute Osiezagha,
  • Bhupesh Kumar Mishra,
  • Temitayo Matthew Fagbola

摘要

This research explores the application of predictive analytics through Machine Learning (ML) algorithms to enhance Mobile Network Key Performance Indicators (KPIs), specifically focusing on Reference Signal Received Power (RSRP) as coverage and Reference Signal Received Quality (RSRQ) as quality. Various regression and classification modelling techniques were applied to drive-test measurements collected around the University of Hull, utilizing supervised ML algorithms such as Decision Tree (DT), Logistic Regression (LogisticR), Random Forest (RF), Support Vector Machine/Regressor (SVM/SVR), Light Gradient Boosting Machine (LightGBM), K-Nearest Neighbour (KNN), Extra Trees (ET), Extreme Gradient Boosting (XGB), Multi-Layer Perceptron (MLP), Deep Neural Network (DNN), Gaussian Naïve Bayes (GNB), and Gradient Boosting (GB) to benchmark the performance of four Mobile Network Operators (MNOs)/Mobile Virtual Network Operators (MVNOs) at various locations around the University of Hull, with additional model validation conducted in Hull City Centre, Barton Upon Humber, and Newland as use cases. The Random Forest (RF) model emerged as the best-performing algorithm, achieving a Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) below 3.7, a Mean Absolute Percentage Error (MAPE) under 7.03, a Coefficient of Determination (R2) greater than 74%, a Receiver Operating Characteristic Area Under the Curve (ROC_AUC) above 93%, and an Accuracy exceeding 82%. Additionally, the ensemble learning (EL) model, which combined the strengths of RF, GB, ET, SVR, XGB, and LightGBM for regression, and LogisticR, SVM, MLP, GB, ET, and RF for classification, delivered an overall performance with RMSE and MAE below 4, R2 above 72%, accuracy exceeding 81%, and ROC_AUC above 85%. This highlights the EL model’s ability to predict network coverage (RSRP) and quality (RSRQ) as excellent, good, fair, bad, or poor with high precision. This study demonstrates the uniqueness of integrating multiple KPIs (RSRP and RSRQ) and prediction techniques (regression and classification) within an Artificial Intelligence (AI)-driven solution, providing a robust framework for improving network performance, particularly in scenarios where data collection through drive testing is limited.