Artificial neural network modeling of rare earth element solvent extraction based on pH and extractant concentration

The 15 lanthanides, along with scandium and yttrium, are considered rare earth elements (REEs). Most minerals containing REEs typically include 9 to 10 different rare earth elements. The most prevalent method for separating REEs is solvent extraction, which depends on the distribution coefficient, extractant type, pH, extractant concentration, and temperature. Predicting the distribution coefficient of solvent extraction of rare earth elements remains a challenging task due to the complex interactions between chemical species. Conventional regression-based models are failing dramatically due to different extractants and conditions. This study presents two artificial neural network (ANN) models trained to predict the distribution coefficient (log D) of REEs based on (1) pH and (2) log-scaled extractant concentration. Experimental data were preprocessed using one-hot encoding for extractant identity and augmented with atomic number as a proxy for lanthanide characteristics. The ANN architectures demonstrated strong predictive performance with a high correlation between predicted and experimental log D values. This work contributes to data-driven modeling in hydrometallurgy and supports the integration of neural networks into process analysis and optimization of REE extraction systems.