Prediction of Telecommunication Network Outage Time Using Multilayer Perceptron Modelling Approach

Abstract

As the demand in communication traffic grows; there should be the provision of reliable telecommunication network to users. Operators however, face a myriad of challenges in fulfilling their part of the contract such as network outage. The phenomenon of network outage has been a challenge that every network operator is consistently trying to avoid. In this study, a multilayer feedforward neural network also called multilayer perceptron (MLP) was adopted to model network outage time of network elements or systems. The MLP network was trained on a 150 samples of daily network outage time data obtained from the Network Operating Centre of an operator in Ghana. The data covered a period of 1st January to 30th May 2018 and was analysed with Matlab software. In developing the model, the input and output layers were kept constant, while the number of neurons were varied from 1 to 20 to obtain a good prediction. The performance of the models were measured by the Mean Squared Error (MSE), Root Mean Squared Error (RMSE) and the Correlation Coefficient (R). After careful and extensive training, validation and testing, 20 models were developed. The MLP selected was 1-4-1 which produced MSE, RMSE, and an R-value of 0.0000024321, 0.00160 and 0.99993 respectively with a prediction accuracy of 97.5%. The study concludes that downtime prediction can be improved by feed forward neural network optimized using Levenberg-Marquardt with sigmoid function in the hidden and linear activation function in the output layer.