Machine learning for predicting and optimizing the CO2 uptake in porous organic polymers

Porous organic polymers (POPs) are promising materials for carbon capture due to their high surface area, tunable porosity, and structural versatility. In this study, a machine learning (ML) framework was developed to predict and optimize the CO₂ adsorption capacity of POPs using structural and process-related parameters. A dataset of 190 entries was compiled from experimental literature, incorporating features such as surface area, pore volume, porosity, temperature, and pressure. Five ML models—Random Forest (RF), Gradient Boosting (GB), Support Vector Regression (SVR), Artificial Neural Network (ANN), and a hybrid RF+GB ensemble—were trained and evaluated using 5-fold cross-validation and grid search for hyperparameter tuning. The Gradient Boosting model exhibited the highest performance, with R² = 0.963, MAE = 0.166, and MAPE = 15.39 % on the test set. Feature importance analysis revealed that pressure and temperature were the most influential factors, while surface area also contributed significantly to predictive accuracy. To further enhance CO₂ uptake, a Genetic Algorithm (GA) was integrated with the ML model to identify optimal material properties and operating conditions. The GA-optimized system predicted maximum uptakes of 4.5 mmol/g at 273 K and 3.2 mmol/g at 298 K. This integrated ML-GA approach demonstrates a powerful tool for guiding the rational design of high-performance POPs and optimizing carbon capture conditions. It enables efficient screening of material candidates and operating scenarios, supporting accelerated development of adsorption-based CO₂ capture technologies.