Artificial Intelligence-Enhanced CO2 Capture Capacity Predictions of Functional Ionic Liquids Based on Group Contribution Descriptors

Ionic liquids (ILs), especially functional ILs, have attracted much attention as tunable sorbents for CO2 capture owing to their unique properties. Given the significant role of the structure, an extensive data bank was established containing 2500 experimental capacity data points of CO2 capture as a function of operational temperature and pressure by 232 kinds of ILs during 2002 to 2025. These total data were randomly divided into an 80% training set and 20% test set. Based on the group contribution (GC), the structures of these ILs are divided into 44 types of fragments, including ionic fragments (IFs). All 44 fragments, T, and P were used as the input parameters, while the capacity was regarded as the output parameter to establish four kinds of GC-based machine learning (ML) regression models, multilayer perceptron (MLP), support vector regression (SVR), random forest (RF), and gradient boosting regression (GBR), representing a good balance between complexity and performance. Among these models, the GC-GBR model demonstrated the strongest predictive ability. Feature importance and SHAP analysis uncover a quantitative structure–property relationship (QSPR) where abundant aliphatic groups are strong indicators of low performance, while the amine group is identified as a key promoter of high performance. This interpretability transforms the model from a black box into an interpretable tool for molecular design. This is the first time that GBR-SHAP has been used for analysis of CO2 capture performance by functional ILs, and we hope that these GC-based ML models could be used to develop functional ILs in the future for efficient CO2 capture.