Validating the goodness-of-fit of machine learning models through backward models

In recent years, machine learning (ML) approaches have gained significant attention in subsurface property estimation problems. However, due to the data-driven nature, people often interrogate the reliability of such ML approaches, especially for regions where ground truth data are sparse. In this study, in the application of subsurface property estimation, we investigate the feasibility to evaluate the goodness-of-fit of ML predictions by using a backward ML model. We find that such backward ML models provides an alternative way to qualitatively evaluate the ML prediction results in case ground truth data are not available.