A Study on Urban Structure Map Extraction for Radio Propagation Prediction using XGBoost

Recently, the rapid increase in mobile data traffic and the diversification of wireless communication services have led to demand for the development of high-quality mobile service areas. Therefore, the modeling of complex radio propagation characteristics in the practical communication environment is an important issue. Radio propagation prediction methods based on machine learning have been proposed, but no study on the appropriate range of urban structure maps as features has been conducted. In this paper, we clarify the appropriate extraction range of map data, applying XGBoost to the machine learning algorithm. Moreover, since XGBoost can output the importance of the input features, we can extract only those features that are useful for training and input them to machine learning. This approach is expected to improve both prediction accuracy and speed. The evaluation using measurement data obtained in urban areas showed that the computation time for learning can be reduced by about 80% while improving the prediction error.