Development of aGeneralizable Gradient Boosting Regression-BasedFramework for Predicting the Properties of Single and Binary IL Systems

Ionic Liquids (ILs) have become ubiquitous across various applications, including chemical synthesis, carbon capture, and separation processes. The ability to fine-tune the thermophysical properties of ILs by judiciously selecting cation and anion combinations in conjunction with their favorable properties makes them incredibly versatile materials. Recently, binary IL systems have gained particular interest as electrolytes for electrochemical energy storage applications including batteries, supercapacitors, and fuel cells. However, the vast design space of ILs and binary IL mixtures makes experimental characterization of their properties both time-consuming and expensive, thus calling for the development of computer-aided molecular design approaches to overcome these challenges. Herein, we introduce a Gradient Boosting Regression (GBR)-based framework for predicting the density and viscosity properties of single IL and binary IL systems using features extracted from COSMO-RS-generated sigma profiles. We demonstrate the ability of our framework to make generalized density and viscosity predictions of single and binary IL systems previously unseen by the model using the same set of descriptive features. We further demonstrate an approach for improving single IL viscosity predictions using the GBR framework trained on binary IL data. In summary, this work presents a framework for predicting IL properties of single and binary IL systems and provides important insights for the improved computer-aided design of IL systems.