Abstract:
Mobile Network Operators(MNOs) are expected to wisely and proactively man age the required QoS of the ever increasing cellular network traffic for the diverse
services that are expected to be provided over their networks. New radio resource
dimensioning, continuous expansion and optimization, and making business strate gic decision requires knowledge of the amount and mix of traffic going to be carried
over the mobile network. In Ethiopia, Ethio Telecom have been the only integrated
telecom service provider aspiring to meet the tremendously growing cellular traf fic demand in the country. The activities performed by the company to meet
the demands of its customers is being challenged by unexpected explosive growth
in traffic demand and evolution in user behavior. In a multi-operator market the
company have to stay alert and monitor the service demand in advance in order to
stay competitive both in terms of meeting its customers’ Quality of Service(QoS)
requirement and the return on investment (ROI).
This thesis proposed a cellular network forecasting model using state of the art
deep learning technique, LSTM-RNN to predict the future peak hour traffic volume
and mix that will be carried over the mobile network. This is achieved by train ing and validating the model using both CS (Circuit Switched) and PS (Packet
Switched) peak hour traffic data collected from Ethio Telecom mobile network.
The proposed traffic forecasting model can be used as a tool to predict future
demand which will be an input for cellular radio network design, optimization and
to make high level strategic decision on service and technology evolution.
The model is developed to perform prediction for voice, downlink and Uplink
data traffic. Model performance is evaluated using 10% of the dataset which is
a prediction about three months’ time ahead and the mean absolute percentage
error (MAPE) is found out to be 0.07%, 0.03% and 0.04% for voice, downlink and
uplink data traffic respectively.
