Prediction of CO2 Solubility in Ionic Liquids Based on Machine Learning and Analysis of SHAP

Abstract

Listen

Translate article

The combination forms of anions and cations in Ionic liquids (ILs) which was solvent for the absorption of CO2 were extremely numerous. Consequently, a Machine Learning (ML) model of Transformer was used to measure the solubility of CO2 in each new ILs in this study. The model used 8869 data points and encoding anions and cations based on SMILES. The r, R2, RMSE and MAE were used as error indicators. Additionally, a decoding method was referenced for the first time in the field of ML predictions of CO2 solubility in ILs, which improved the data processing results. As a result, the model achieved better predictive standards. SMILES based SHAP analysis was used on the model to understand the black box operation. The results of the SHAP analysis identified the structural factors that influenced the model's results in the solubility of CO2 in ILs.