Artificial Neural Network Based Prediction of Key Performance Indicators for Mobile Telecommunications

Abstract

This paper presents comparativestudies among artificial neural network neurons. Four key performance indicators were predicted using neural network. The key performance indicators and weather parameters for Osun State University, Osogbo, Nigeria were employed. MATLAB R2020a was employed to develop the neural network models. Three different neural network models were developed. Model A, Model B and Model C with ten neurons, fifteen neurons and twenty neurons respectively, the hidden layer of the models was Log-sigmoid activation function, and the linear activation was used at the output layer of the models. The three models were compared using mean absolute error and mean square error. The best performing model was Model B with fifteen neurons. Its mean absolute error and mean square error is 0.0909 and 0.0123 respectively. Model A with ten neurons was the least performing model with mean absolute error and mean square error of 0.0990 and 0.0148 respectively. The results show that for a model to be robust, several neurons should be tested to establish the most effective model.