Performance Evaluation of GeoAI-Based Approach for Path Loss Prediction in Cellular Communication Networks

Accurate signal path loss models for predictions are crucial in current cellular communication networks. Recently, numerous path loss estimation methods have been presented to improve the efficiency of networks. However, most of these existing models do not include spatial data such as land use/land cover, terrain elevation, building height, and the effect of topography. To address this issue, this study proposes a GeoAI-based technique for path loss estimation in cellular communication networks, addressing existing models’ lack of spatial data integration. Support Vector Regression, K-Nearest Neighbor, Random Forest, and multi-layer perceptron (MLP) artificial neural network models are evaluated using field measurements in an urban, suburban area in Van, Turkey, across various frequencies. Among the models, MLP with three hidden layers, nine input variables, hyperbolic tangent activation function, and Adam optimization method performs best. At 900 MHz, MLP has been observed with MSE, RMSE, MAE, and R values of 0.22 dB, 0.47 dB, 0.46 dB, and 0.99 dB, respectively. Lastly, a comparison of the developed model to the Free space, COST 231, Ericsson, and SUI models revealed that the GeoAI-based path loss models outperformed the empirical models regarding prediction accuracy and generalization. This study underscores the significance of integrating spatial data into path loss prediction, particularly in diverse urban and suburban environments, for optimizing cellular communication networks.