From Spatial Urban Site Data to Path Loss Prediction: An Ensemble Learning Approach
Sotirios P. Sotiroudis, Achilles D. Boursianis, Sotirios K. Goudos, Katherine Siakavara
Abstract
Machine learning (ML) models have become increasingly popular in the field of path loss prediction. Their performance depends profoundly on the data they use as their input. The work at hand proposes and evaluates new input features for urban propagation. These features were obtained via image processing tools. Moreover, we propose a new two-level ensemble model using the concept of stacked generalization. The proposed model is based on seven different basic ML models. To the best of the authors knowledge, this is the first time that a two-level ensemble method is applied to a modeling problem in electromagnetics. The results demonstrate that the new input features, coupled with the new ensemble, provide improved prediction performance.