Machine Learning Prediction of Density of Ionic Liquid–Organic Solvent Systems for Green Chemistry Applications: Ensemble Model, Interpretability, and Validation

Due to their tunable physicochemical properties, ionic liquid–organic solvent systems are increasingly important in green chemistry and energy materials. This study aims to accurately predict the density of such systems using machine learning. Four models were developed and compared: multilayer perceptron, random forest, convolutional neural network, and ensemble model. A curated data set of 15,826 data points covering 83 ionic liquids and 30 organic solvents was used for training and testing. The ensemble model outperformed others, achieving an R 2 of 0.9945, MAE of 0.0063, MSE of 0.0001, and RMSE of 0.0113 on the testing set. Bayesian optimization was applied for hyperparameter tuning, and SHAP analysis was used to interpret the feature importance. Model robustness was verified via applicability domain analysis and learning curves. Validation on an independent data set confirmed the ensemble model’s high predictive accuracy, providing a reliable tool for the design and application of ionic liquid–organic solvent systems. In addition, this study also conducted a greenness ranking of ionic liquids and organic solvents, providing an important reference for their green selection and replacement.