Predicting Viscosity of Ionic Liquids Using Structure-Based Models

Due to the tunability of ILs' structure, there are many ion pairs with a wide range of viscosity. Thus, machine learning methods are used to efficiently predict the viscosity of ILs and establish accurate structure-property relationships. Structure-based models were built to predict dynamic viscosity in ionic liquids (ILs) using two approaches: an artificial neural network (ANN) with clustering and dimensionality reduction along with a gradient boosting method, CATBoost, with full descriptor utilization. Using a cheminformatics software, RDKit, to generate descriptors, both models demonstrate strong generalizability across the viscosity range and variety of ILs. For the ANN approach, data is categorized into four distinct groups using the KMeans clustering algorithm along with a balancing function, and optimal hyperparameters, such as the number of clusters and principal components, are determined using the elbow methodology. Comparative analysis shows that the ANN with dimensionality reduction achieves R 2 values of 0.94 and 0.87 for the training and test sets, respectively, while CATBoost with full descriptors yields R 2 values of 0.94 and 0.85, demonstrating that both approaches provide robust prediction capabilities. The generality of the method provides a wide range of applications which can be extended to the prediction of various properties of soft materials based on their structures.