In this paper an ensemble of 1,000 deep neural networks were developed to quantify uncertainty in predicting trace-element concentrations in geochemical datasets, particularly when dealing with very small and highly imbalanced data distributions. The models use concentrations of 11 major elements and pH as input parameters to predict trace element conentrations.
This tool closes data gaps in existing hydrochemical datasets and reduces analytical costs for long-term monitoring. Learn more on our Publications page!
