A Machine Learning Approach to Sleep Mode in Dense Cellular Networks
Abstract
The explosion of data hungry applications has led to a rapidly increasing demand for cellular
network services. Mobile operators are responding by increasing the number of Base Stations
(BSs) in their networks to enhance capacity. The densi cation of cellular networks however,
comes with extensive capital and operational investment due to the increase in power consumption
of the networks. In addition, this is raising environmental concerns over increase
in greenhouse gas emissions. Moreover, current cellular networks are designed to handle peak
traffic to other the best Quality of Service (QoS) at all times and yet network subscribers are
not uniformly distributed making the 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. Three machine learning (ML) algorithms,
Random Forest (RF), Support Vector Regression (SVR) and Polynomial Regression (PR) were
used to carry out traffic prediction and then compared. Results obtained from the different
algorithms were analyzed using three performance metrics; accuracy, Root Mean Square Error
(RMSE) and Mean Absolute Error (MAE). RF was selected since it gave the best results with
84% accuracy, and as such, the traffic pro les from this algorithm were adopted for sleep mode
implementation on a homogeneous network. Simulation results show that the strategic sleep
mode scheme performs best over a 24-hour period with a 50.12% energy saving gain over the
conventional scheme and 15.55% energy saving gain over the random scheme. In addition, the
strategic scheme achieves a 53:4 kW/km2 over a 24-hour period, which proves that ML traffic
prediction-based sleep modes are instrumental in achieving energy efficient cellular networks.