Channel Loss Prediction For Lte At 1800 Mhz Using Classical and Machine Learning-Based Models

Wireless communication systems--based on Long Term Evolution (LTE) technology--require accurate channel loss prediction to support network planning and performance analysis. This paper addresses channel loss prediction for LTE at 1800 MHz using traditional techniques and Machine Learning approaches, to do that, this work is based on a dataset of measurements of received signal strength indicators and transmitter–receiver distances, from which a common support dataset is constructed. Three experimental scenarios are considered: the use of the complete dataset, the removal of outliers, and the combination of outlier removal with data augmentation. For each scenario, channel losses are computed using both mean- and median based estimates, and model performance is evaluated using the Root Mean Square Error (RMSE). The results show that the application of outlier removal and data augmentation consistently improves the accuracy of the model across all approaches. Among traditional methods, logarithmic curve fitting achieves the best performance, with an RMSE of 5.0410. Machine Learning models based on Gaussian Process Regression provide the lowest prediction error, with the Rational Quadratic kernel achieving an RMSE of 4.8046. These results confirm the effectiveness of data preprocessing strategies and data-driven regression models for LTE path loss prediction, while highlighting the advantages of Gaussian Process–based methods in capturing the underlying large-scale propagation behavior.