An ensemble machine learning approach for enhanced path loss predictions for 4G LTE wireless networks

Abstract

SummaryAccurate path loss prediction models are indispensable in modern wireless communication systems. In recent times, several path loss prediction models have been proposed to improve network performance. However, most of these models have not addressed the fundamental issues. The problem of deploying a single path loss prediction model that fits well in all wireless propagation environments remains. To address this problem, we present machine learning‐based ensemble methods to path loss predictions. Specifically, ensemble methods have been introduced to improve signal prediction accuracy and performance. Additionally, the radial basis function (RBF) and multilayer perceptron (MLP) neural network models were deployed and improved by adding more network parameters to their respective input layers. Results revealed that the RBF model with an increase in the number of centroids reduces the mean square error (MSE). Also, the Gaussian kernel function gives lower MSE compared to the multiquadric and inverse multiquadric functions. The bagging and blending ensemble path loss prediction models were introduced for optimal performance. An RBF neural network of different clusters was incorporated into the bagging algorithm as base learners. The RBF, MLP, blending, and bagging were examined using standard metrics for the training, testing, and validation dataset for model validation. The developed bagging ensemble path loss prediction model gave the lowest errors (MSE = 0.0011 dB, SSE = 0.6069 dB, MAE = 0.0245 dB, &R = 0.7484 dB) for the datasets. The bagging ensemble method acts as a variance and error reduction mechanism because it predicted path loss closest to measured data and is suitable for near precise path loss predictions.