Neural Network based 3G Mobile Sites Fault Prediction: A Case Study in Addis Ababa, Ethiopia
No Thumbnail Available
Date
2018-11
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
AAU
Abstract
As cellular mobile networks are evolving in technology and service type, the number of
mobile network site infrastructure is increasing. Nowadays, faulty cellular mobile
network sites per day in ethio telecom are significant in number and have big impact to
customers and operator in QoS, revenue and maintenance cost. Fault maintenance
techniques commonly applied by ethio telecom is corrective maintenance approach. This
only helps to recover services after interruption. However, it is important to implement
the proactive maintenance approach to make mobile sites reliable and available. This
helps to provide services according to standards and improve the quality of service
delivery to customers. To mitigate mobile network site faults before happening, fault
occurrence time prediction is an important technique for the implementation of proactive
maintenance strategy. This Neural Network based 3G Mobile Fault Occurrence
Prediction research work is conducted based on the Nonlinear Auto regressive (NAR)
Neural Network time series prediction method using Addis Ababa 3G mobile sites in a
case study. To train the neural network 15,950 actual fault occurrence time data are used.
The algorithm used to train the neural network is Levenberg-Marquardt which is fast and
efficient, and an iterative approach of hidden layer neuron number selection is applied.
Finally, the best model is selected with minimum value of mean square error of
prediction. Also, the model is tested with actual fault occurrence time which was not used
in the training and achieved 90.71% in prediction. Therefore, it is efficient in prediction
accuracy, fast and adaptive with future data.
Description
Keywords
Mobile Site, Fault, Corrective Maintenance, Proactive Maintenance, Fault Occurrence Time, Prediction, Neural Network, Time Series