Prediction of Carbon Dioxide Adsorption Performance of Covalent Organic Frameworks (COFs) Based on Machine Learning: Identification of Key Factors and Model Deployment.

Covalent organic frameworks (COFs) have recently garnered significant recognition as environmentally friendly, high-performance adsorbents with exceptional greenhouse gas sequestration capabilities. This study systematically predicts the CO2 adsorption performance of COFs by integrating six machine learning models, including XGBoost and Random Forest, etc., and interprets key influencing factors using SHAP analysis. As the best model, XGBoost's performance on the test set can reach an R 2 of 0.9504, an MAE of 0.3129, and an RMSE of 0.5157, SHAP analysis shows that pressure is the most critical factor affecting the CO2 adsorption performance of COFs (accounting for about 38.8%). Furthermore, a graphical user interface (GUI) was developed to facilitate the direct utilization of the model for predicting CO2 adsorption performance in COF materials. This study presents a 3-fold contribution: (1) A systematic evaluation of six ML algorithms establishing XGBoost as the optimal predictor; (2) SHAP analysis providing a model-based interpretation of dominant factors and their nonlinear contribution trends; and (3) a deployable GUI was developed to facilitate user-friendly prediction of CO2 adsorption performance under reported descriptor and condition inputs.