Predicting thermophysical properties of dialkylimidazolium ionic liquids from sigma profiles

We present a Support Vector Regression (SVR) machine learning framework for predicting the viscosity, ionic conductivity, and density of imidazolium ionic liquids (ILs) using a universal set of features extracted from COSMO-RS sigma profiles. To train and test the SVR model, we assembled three property datasets with approximately 40 different ILs, each consisting of over 1000 experimental datapoints measured across a wide range of temperatures and pressures. From calculated sigma profiles we extract IL descriptors or features that are readily fit by using the SVR model. After cleaning of the measurement datasets and selecting these IL features, we compare the performance of the radial basis function (RBF) and linear kernels using a standard k-fold cross-validation to separate the respective datasets into training and testing datasets without bias. Using these results, we demonstrate the ability of the RBF-SVR model to predict the viscosity, conductivity, and density of unobserved ILs at atmospheric pressure. (C) 2021 Published by Elsevier B.V.

Automated summary

This study introduces an SVR machine learning framework for predicting properties of imidazolium ionic liquids by extracting universal features from COSMO-RS sigma profiles. After training and testing the model, it shows the ability to predict viscosity, conductivity, and density of unobserved ILs using the selected features. The performance of the model is compared with different kernels through cross-validation to ensure unbiased results.