A deep learning approach to traffic prediction in emergency situations: the case of COVID-19

Abstract

Communication in a very important part of our lives and technological advances are being made each day to improve on the state of communication. Many services such as commerce, medication, academia among others are now facilitated by technology and this requires a good communication between these services using the respective devices. According to cisco,the number of devices connected to the internet was projected to reach about 28.5 billion in 2022 from 18 billion in 2017[1]. This calls for efficiency of the networks set up by the telecommunication companies since they facilitate most of communication. Various steps have been made in ensuring the efficiency and resilience of these networks through network optimization methods like base station sleeping[2][3], network traffic prediction, self diagnosis, self healing[6] among others but in all these, Cellular network prediction plays a crucial role in most of those optimization techniques since it helps the network works to plan ahead. Advanced techniques have been suggested to increase the accuracy of these predictions and some of these use models based on Long-Short Term Memory(LSTM), Convolutional Neural Network-LSTM, Spatial–Temporal Cross-domain neural Network (STCNet) among others. The issue with these models is that they do not cater for situations where there is a very significant change in the network usage patterns and volumes of cellular traffic used.In addition, they have a short forecasting horizon of about 6 hours, maximum. In this report, we developed a model that is able to do cellular network prediction in all situations and and also forecast up to 32 hours into the future. We used the first Covid-19 season in Uganda as our case study