dc.description.abstract | Current cellular networks are designed to handle peak traffic so as to offer the best qual ity of service at all times and yet network subscribers are not uniformly distributed making
traffic within the network vary spatiotemporarily. This results into subsequent energy losses
throughout the day, which negatively affect the energy performance of the overall network.
This research therefore, aims to propose a machine learning approach to sleep modes in future
cellular networks.
The data used in this work is 3G packet-switched traffic data collected from five neighboring
Base Stations in Kampala, Uganda. It was collected from a major Telecommunications Oper ator for a period of sixteen weeks spanning from 1st December 2019 to 21st March 2020. The
scope of our research is limited to the temporal variations of this traffic and was found to have
variations during the analysis with lower traffic levels in the morning hours and higher ones in
the evenings.
Three machine learning algorithms, random forest regression, Support Vector Machine and
polynomial regression were used to carry out traffic predictions basing on the traffic profiles.
Results obtained from the different algorithms were analyzed using parameters such as accuracy
levels, root mean square error and mean absolute error. Random forest was chosen since it gave
the best results of the performance metrics (84% accuracy), and as such, the traffic profiles from
this approach were adopted for sleep mode implementation.
A homogenous cellular network was simulated using MATLAB starting with 400 base stations
and 800 users. Three sleep mode schemes; strategic, random and conventional were applied
on to the network and our results showed that the strategic sleep mode performs best over
the twenty-four-hour period with a 50.12% energy saving gain over the conventional scheme
and 15.55% energy saving gain over the random scheme. In conclusion, machine learning
traffic prediction-based sleep modes are effective in making cellular networks energy efficient.
In future, more research can be done in using deep learning models for traffic prediction to
achieve greater efficiency | en_US |