Building Univariate Time Series Forecasting Models for Network Traffic by Evaluating Statistical and Machine Leaning Techniques: The Case of Ethio Telecom Broadband VSAT Hub Networks
No Thumbnail Available
Date
2022-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Network traffic congestion is the major challenge in telecom service providers
where usually they use a limited resource to deliver the services to their customers.
That leads to network performance and Quality of Services (QoS)
degradation so that do not meet customer satisfaction. The Very Small Aperture
Terminal (VSAT) network in Ethio Telecom delivers broadband services
through satellite with a limited capacity. Therefore, this thesis aims to
study the VSAT network traffic patterns to propose the traffic forecasting
model. That will be used as a solution to enhance the network resources
based on the prediction of the future traffic demand and as input for network
planning and optimization works.
In this study, the VSAT data traffic recorded for one year from 01-Mar-2020
to 28-Feb-2021 was collected. The dataset is used for data preprocessing,
statistical analysis, model training and testing. All the tasks are performed
by Python software. In addition, existing time series forecasting methods
are selected from statistical and machine learning models that includes the
Exponential Smoothing Methods (ESM), Autoregressive Integrated Moving
Averages (ARIMA), Seasonal ARIMA (SARIMA), Artificial Neural Network
(ANN) variants Multilayer Perceptron (MLP), Recurrent Neural Network
(RNN) and Long Short Term Memory (LSTM). The forecasting accuracy
metric of Root Mean Square Error (RMSE), Mean Absolute Error (MAE)
and Mean Absolute Percentage Error (MAPE) are used to evaluate the
forecasting performance of the models. These are applied to examine and
choose the model having the minimum forecasting errors. As a result, the
RNN model is identified the best model and improved the forecasting performance
by 44.94% than the Triple Exponential Smoothing (TES) model which is a variant of ESM. Therefore, the RNN model is proposed to Ethio
Telecom for future network planning and optimization to VSAT networks.
Description
Keywords
VSAT, Network data traffic, Forecasting, ESM, ARIMA, SARIMA, ANN