Spatiotemporal Mobile Data Traffic Prediction Using Convolutional Long Short-Term Memory: The case of Addis Ababa, Ethiopia
No Thumbnail Available
Date
2019-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Globally, exponential data growth is observed with mobile traffic generated from devices like
tablets, smartphones and other devices. Likewise, Addis Ababa city’s cellular network data traffic
is increasing exponentially. To absorb this high traffic demand ethio telecom, the telecom service
provider in the city continuously expands and optimizes the cellular network. Having knowledge
of the growing data traffic demand in advance at a given time and space will assist ethio telecom’s
planning strategy and optimization.
Recently, few studies are conducted to forecast Addis Ababa city’s Universal Mobile
Telecommunication System (UMTS) network traffic using statistical time series models and neural
network models. However, the studies deal with only time-domain forecasting and recommend
to do from a spatial point of view. Moreover, another study modeled the spatiotemporal mobile
data traffic, which can capture the space and time variation of UMTS data traffic in the city; the
study recommends the need of spatiotemporal data traffic prediction.
In this thesis, a deep neural network model, specifically Convolutional Long Short-Term Memory
(ConvLSTM), is used for spatiotemporal data traffic demand prediction of Addis Ababa city. Three
months' real dataset from 739 base stations is collected and preprocessed from ethio telecom’s
UMTS network. After defining geographical grids, the ConvLSTM model is applied, which can
capture spatial correlations through convolution operators and temporal dynamics through the
LSTM network for prediction.
The proposed model can predict up to six hours of future data traffic with a root mean square
error (RMSE) of 1.37. Additionally, the predicted data traffic demand is analyzed with respect to
blocked data traffic at a given space and time which gives significant insight to the optimization
processes like load balancing.
Description
Keywords
Spatiotemporal, data traffic prediction, UMTS network, deep neural network, ConvLSTM