Metals are key components of modern devices; however,
available
resources of these metals are limited. In this study, we used machine
learning (ML) to curate ionic liquids (ILs) that are suitable for
metal extraction. We proposed classification and regression models
to unravel hidden patterns between IL structures and their specific
properties, i.e., metal selectivity and eco-toxicity. Evaluations
of ML models using cross-validation indicate that the models were
reliable, as described by the accuracy score (0.82) and R
2 value (0.76). The models also revealed that the metal
selectivity of ILs was determined by the cation and anion structures,
and the eco-toxicity level was primarily affected by the cation structures.
Guided by predictions from the trained models, we selected three ILs
(out of the 150 IL structures we initially proposed) that have extraction
selectivity toward platinum, lithium, and neodymium as well as low
eco-toxicity. We then prepared the ILs in the laboratory and assessed
their performance by standard solvent extraction. The experiments
indicate that the recommended ILs from ML could selectively extract
the targeted metals with high extraction efficiency (>80%), which
demonstrates the feasibility of ML as a promising toolkit that can
help accelerate innovations in metal extraction.