Continuous Bayesian Networks vs. Other Methods for Regression in Environmental Modelling

Abstract

Several algorithms have been developed in the literature to solve regression problems. We propose a novel methodology based on Bayesian networks (BNs) to deal with regression problems in environmental research. To demonstrate its capabilities and strength, we compare a BN model with 3 other methods commonly used to solve regression tasks, in terms of their root mean squared error (RMSE). The errors were depicted on error maps, providing information about the reliability of the predictions in each observation. The results show that BNs are competitive with other popular methods.