Uncertainty quantification of geochemical data imputation using Monte Carlo dropout

Machine learning models have shown their promise in geochemical data imputation tasks. However, being blackbox solvers, these models require more confidence in their predictions. Using uncertainty quantification methods for deep neural networks can increase the reliability of their predictions. In this paper, we use Monte Carlo Dropout to estimate uncertainty in geochemical data imputation. Multiple forward passes with different dropout configurations yield a predictive distribution for the unknown analytes. The mean of this distribution is used as the prediction, while the standard deviation expresses the uncertainty of the neural networks. Two different scenarios, namely the WACHEM and WAMEX databases containing multi-element geochemical data for rock samples, illustrate the predictive accuracy of the method and its capability to measure the associated uncertainty. Dropout values of 0.1-0.2 were identified as a good balance in prediction accuracy and model uncertainty.