A quantitative prediction of the viscosity of ionic liquids using S_σ-profilemolecular descriptors

Abstract

In this study, two novel QSPR models have been developed to predict the viscosity of ionic liquids (ILs) using multiple linear regression (MLR) and support vector machine (SVM) algorithms based on Conductor-like Screening Model for Real Solvents (COSMO-RS) molecular descriptors (Sσ-profile). A total data set of 1502 experimental viscosity data points under a wide range of temperatures and pressures for 89 ILs, is employed to train and verify the models. The Average Absolute Relative Deviation (AARD) values of the total data set of the MLR and SVM are 10.68% and 6.58%, respectively. The results show that both the MLR and SVM models can predict the viscosity of ILs, and the performance of the nonlinear model developed using the SVM is superior to the linear model (MLR). Furthermore, the derived models also can throw some light onto structural characteristics that are related to the viscosity of ILs.