Study on the Efficiency of Models Forecasting the Load on the Servers of a Cellular Operator

The problem of predicting the possible loads in a cellular network operation can be reduced to building a forecast on the possible number of calls directed to one gateway (PGW) within the given period of time. Having these data for all gateways in the network, it is possible to organize the optimal distribution of resources, prevent overloading of the gateways and, as a result, failures in the entire network operation. A statistical analysis of actual data collected by automated measuring systems on the nodes of a mobile network is carried out and the most suitable data for building forecasting models are identified. The results of the research on the possibility and effectiveness of the application of the mathematical models realized in constructing such a forecast by using machine learning methods such as linear regression, k-nearest neighbors (KNN), and random forest are presented. It is established that in order to solve the problem of building a short-term forecast on the number of requests that are to enter the server, it is not necessary to use complex models that require computing resources. Based on the calculated quality metrics, it is found that the most accurate forecast can be obtained by using a linear regression model.