This study presents the development of an Ensemble Neural Network (ENN) model for real-time path loss prediction and 5G network coverage optimization in the Trans-Amadi industrial layout of Port Harcourt, Nigeria. The study uses the drive-test data on different seasonal conditions taking both normal, Harmattan, and rainy season conditions and compares the achieved results of using the ENN model with the established empirical models of Okumura, COST-231 Hata, and Log-distance models in the region. A hybrid scheme of optimization of ENN was employed that comprised both Bayesian Regularization and Adam optimizer to optimize the generalization as well as convergence. The model was evaluated and tested using the key performance measures such as Mean Squared Error (MSE), R-squared (R 2 ) and prediction accuracy. It is found out in the experimentation that the ENN model has a high precision of predictions, 98% with low deviation (2.00% -2.14%) with reference to real field measurements, at whatever weather in the continuum. Comparatively, the deviation rates were remarkably higher with the traditional models, which proves that ENN model is much more efficient in modelling the non-linear propagation behavior due to influence of environmental dynamics. The model has also been coupled with a Minimum Cost Resource Allocation (MCRA) algorithm in order to allow dynamical network optimization and spectrum efficient use. The results indicate the ENN model as a reliable and scalable prediction of path loss in 5G deployment especially as part of complex urban environments. It can be used in network resource allocation, in such a way that better Quality of Service (QoS), power control, interference reduction, cell planning decisions could be supported. The study highlights the significance of the machine learning-based methods towards the improvement of the performance of wireless communication systems in practice.
A Thu, study studied this question.