PRLGB: Progressive Residual LightGBM for 5G Quality-of-Service Resource Allocation Prediction

<title>Abstract</title> <p>This study presents a focused evaluation of ensemble machine learning models for Quality of Service (QoS) prediction and resource allocation optimization in 5G networks. We compare five tree-based ensemble models: Base LightGBM, Random Forest, XGBoost, CatBoost, and a proposed Progressive Residual LightGBM (PRLGB) on the Kaggle 5G Resource Allocation Dataset. PRLGB implements a three-stage progressive architecture comprising: (1) coarse prediction with baseline LightGBM, (2) residual specialization learning prediction errors, and (3) uncertainty-aware meta-fusion using engineered features \(([f_1(X), f_2(X), |\hat{y}_2(X)|])\). Evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (\((R^2)\)) metrics on the test set, PRLGB achieves MSE = 4.8317, RMSE = 2.1981, MAE = 0.8357, and \((R^2 = 0.9449)\), improving RMSE by 4.52% and MAE by 17.8% over Base LightGBM on this dataset. These results suggest that progressive residual stacking of LightGBM models can provide practical gains for 5G QoS resource allocation prediction on tabular datasets.</p>