New insight into viscosity prediction of imidazolium-based ionic liquids and their mixtures with machine learning models

Ionic liquids (ILs) as eco-friendly solvents have attracted particular attention in various fields of science including the petroleum industry. Among the different families of ILs, imidazolium-based ILs have been the subject of many research studies. However, not enough experimental studies were conducted to determine the viscosity of this family of ILs. Therefore, accurate viscosity prediction is crucial for their practical applications. This study aims to predict the viscosity of imidazolium-based ILs and their mixtures using critical properties of these ILs as input parameters. To achieve this, machine learning (ML) models have been implemented. Furthermore, the performance of these ML models in predicting the viscosity of IL mixtures was compared with a Molecular-based model, ePC-SAFT-FVT (ePC-FVT-MB), and an Ion-based model, ePC-SAFT-FVT (ePC-FVT-MB). Graphical and statistical analyses revealed that the RF model offers the lowest error for viscosity prediction of pure ILs, while CatBoost performs the best for IL mixtures. In addition, sensitivity analysis showed that viscosity decreases with temperature and increases with pressure. The proposed models exhibit high accuracy under varying conditions. Outlier detection using the Leverage method indicated that 95.11% of pure IL viscosity data and 94.92% of mixed ILs viscosity data are statistically valid.