Predicting base station return on investment in the telecommunications industry: Machine-learning approaches

Abstract

Investment in the right location ensures sustainable competition. In the telecommunication sector, the number of base stations (BSs) is one of the most significant investment parameters. When a potential BS is subject to be selected, practitioners will first consider investing in a BS where the return on investment (ROI) is highest. Therefore, the quantifiable objectives are distinctly defined, as it makes sense to choose maximizing features that raise per unit investment. This study provides a solution to evaluate the best BS installation alternative with machine-learning approaches as well as to estimate ROI value by changing the properties that affect the ROI value. For this purpose, the estimation performance of logistic regression, random forest, and XGBoost methods are compared and further strengthened by random forest hyperparameter optimization to provide the best performance. The model, with a success rate of 98.7% according to the $F$-score, showed that it was a robust algorithm. The three most essential features for the ROI value are determined to be voice traffic, data traffic, and frequency cost. These parameters enable a review of the prediction results of telecommunications managers and planning specialists responsible for BS investment.