Machine Learning Approach for Propagation Attenuation Evaluation in 4G LTE Wireless Communication Networks

This study focused on developing signal propagation attenuation models using a machine learning approach. Four commercially installed LTE base-stations in Port Harcourt that operated at 2600 MHz were considered with the extraction of signal data for the study. A field drive-test method was utilized to collect signal data within the environment. The measured signal data were denoised through the rigrsure thresholding method using the wavelet tool in Matlab. The measured denoised propagation attenuation values were estimated using measured unprocessed signal data. The developed hybrid model using the denoised data was designated as the wavelet-GA model, whereas the genetic algorithm (GA) model was used through the unprocessed data designated as the GA model. The RMSE, MAE, and correlation coefficient (R) were used as evaluation metrics to compare the machine learning hybrid Wavelet-GA model with the GA model and the standard COST231-Hata model. The COST231-Hata and GA models were not as predictive as the machine learning hybrid Wavelet-GA model. The machine learning model outperformed the GA and COST231-Hata models in all the base stations in terms of R, RMSE, and MAE values. The corresponding MAE values for base stations 1, 2, 3, and 4 were 1.68 dB, 3.30 dB, 3.02 dB, and 3.34 dB, respectively, while the machine learning hybrid Wavelet-GA model estimated RMSE values of 2.27 dB, 4.61 dB, 3.77 dB, and 3.93 dB. Its high performance was confirmed by the examined R, which showed a strong alignment between the machine learning hybrid Wavelet GA values and the measured propagation attenuation values. However, the COST231-Hata model showed the lowest R and the highest RMSE and MAE values, suggesting a lower degree of accuracy and reliability. The R was also compared with measured propagation attenuation data, and it proved the efficiency of the machine learning model estimated at 94.49%, 84.85%, 92.17% and 93.25% for the base stations, respectively. Validation with data from a different base station confirmed the efficiency of the machine learning propagation attenuation model based on denoised signal data, providing valuable insights for network planning. When evaluated, it showed that the developed machine learning was 97.41% valid within Port Harcourt. Conclusively, it showed that the machine learning propagation attenuation model outperformed existing propagation attenuation models and such recommended for cellular network planning within Port Harcourt as it can remedy the poor quality of service experienced within the areas.