Machine Learning Prediction of Au(III) Extractability of Various Organic Solvents Based on Ion Solvation in Hydrochloric Acid Media

Selective recovery of gold from waste electrical and electronic equipment has attracted increasing attention. Au(III), which is present as tetrachloroauric acid (HAuCl4) in hydrochloric acid media, can be extracted by using various organic solvents, such as ketones and ethers. However, the decisive factors of the solvent for predicting Au(III) extractability have not been confirmed. In the present study, the relationship between the extraction percentage of Au(III) and the properties of the solvent was investigated for 79 types of solvents. Based on the relationships between the Hansen solubility parameters of the solvent and the extraction percentage of Au(III) in 5.0 M HCl with a threshold of 80%, the extractability was classified with 93.7% accuracy. To predict the Au(III) extractability, a machine learning model was constructed using the relationship between the molecular descriptors of the solvents obtained from RDKit as explanatory variables and the extraction percentage of Au(III) as the objective variable. By selection of the descriptors that showed a strong correlation with the Au(III) extraction percentage, the squared correlation coefficient (R2) of the test data for the nonlinear support vector regression model using 35 descriptors was 0.943. Machine learning was performed again with the addition of surface charge density descriptors obtained from the solvent structure using the conductor-like screening model for real solvents. The R2 value of the test data for the nonlinear support vector regression model using 28 descriptors, including the hydrogen bond acceptor descriptor as an additional descriptor, was 0.899, which was slightly lower than that of the model with 35 descriptors. Experimental validation using six additional solvents confirmed the model’s accuracy, with predicted values closely matching the measured extraction percentages.