Performance analysis of regression‐based machine learning models towards intelligent selection of MIMO configurations

Abstract

Fifth Generation (5G) fosters the integration of several technologies for achieving the requirements of rapid and reliable communications. Multiple Input Multiple Output (MIMO) is one such technology implemented in communication systems with the aim of delivering better services to users. Machine Learning (ML) technologies are also being used alongside MIMO since they can aid with the selection of the most appropriate MIMO configuration based on antenna, data, and channel parameters. These parameters can be further used as input features to the models for obtaining the best data rate and throughput. In this paper, three different MIMO configurations, 2 × 2, 3 × 3, and 4 × 4 MIMO using directional and omnidirectional antennas have been simulated. Power propagation from each antenna, MIMO power, channel capacity, data rate, and throughput have been used as response variables. The data obtained from these simulations are used to train regression models for prediction of new data. Frankfurt city is used as scenario in this simulation and the models are analyzed based on the root mean squared error (RMSE) values obtained. The results demonstrate that the interactions linear regression, fine tree and medium tree models provide the lowest RMSE values when predicting the variables for different MIMO configurations using directional and omnidirectional antenna.