Machine Learning Approaches for Further Developing the Understanding of the Property Trends Observed in Protic Ionic Liquid Containing Solvents

Ionic liquid containing solvent systems are candidates for very large compositional space exploration due to the immensity of the possible combination of ions and molecular species. The prediction of key properties of such multicomponent solvent systems plays a vital role in the design and optimization of their structures for specific applications. In this study, we have explored two machine learning algorithms for predicting the surface tension and liquid nanostructure of solvents containing a protic ionic liquid (PIL) with water and excess acid or base present. Machine learning algorithms of multiple linear regression (MLR) and Bayesian regularized artificial neural networks (ANNs) were used to develop semiempirical structure–property models for the data set, which was comprised of 207 surface tension and 80 liquid nanostructure data elements which we previously reported (Phys. Chem. Chem. Phys. 2019, 21, 6810–6827). On the basis of the models, the significance levels for the impact of the alkyl chain length and the presence of hydroxyl groups on cation, type of anion, nonstoichiometry, and presence of water were elucidated. Both models are statistically applicable for designing new PIL containing solvent systems. Furthermore, the generated models were used to create response-surface plots, for both surface tension and liquid nanostructure, interpolated across the compositional space. An additional surface tension data set with 18 new data points within the same compositional space was used to test the prediction ability of models, and the results showed all of the models were successful for prediction. These machine learning approaches are highly suited to the development of structure–property relationships for ionic liquids and particularly for the increasing use of ionic liquid–molecular solvent mixtures.