Machine learning-assisted modeling study on the density and heat capacity of ionic liquid-organic solvent binary systems

Abstract

Properties of ionic liquids (ILs) play a crucial role in solvent design, process design, and process simulation for their various industrial applications. However, ILs come in a wide range of varieties, and obtaining their properties solely through experiments is a time-consuming and expensive process. Therefore, the development of predictive models capable of reliably estimating the properties of these compounds is essential for their effective industrial utilization. In this work, a hybrid model that combines the group contribution method with machine learning algorithms is proposed to predict two important properties (density and heat capacity) of ILs-organic solvent binary systems. This model establishes relationships between the properties and factors such as temperature, pressure, and molecular structure. The predictive performance of three machine learning methods, namely ANN, XGBoost, and LightGBM, is compared using 2504 heat capacity data points and 34,754 density data points, respectively. The results demonstrate that all three algorithms can provide accurate predictions, with the ANN model exhibiting the best performance. For heat capacity, the ANN model achieves a mean absolute error (MAE) of 0.0557 and a squared correlation coefficient (R2) of 0.9952. Regarding density, the ANN model yields a MAE of 0.0049 and an R2 of 0.9942 for the test dataset. Furthermore, the SHAP (Shapley Additive Explanations) method is employed to analyze the importance of each structure and parameter in the prediction results. The analysis reveals that the mole fraction of IL in the binary system has the most significant positive impact on the predicted properties. The models developed in this work not only provide accurate predictions for the properties of binary mixtures but also establish a data-driven framework for predicting the physicochemical properties of multicomponent mixtures.