The growing need for efficient carbon management has intensified interest in amine-based CO 2 capture. In this study, we investigate how critical properties of amines can be leveraged to predict CO 2 capture efficiency in MDEA-based amine solutions. We develop and compare four Machine Learning (ML) and Deep Learning (DL) models namely Bayesian Neural Networks (BNN), Gradient Boosting Neural Networks (GrowNet), Gradient Boosting (GBoost), and the Light Gradient Boosting Machine (LightGBM) using amine critical properties as input features to predict CO₂ solubility in mixed amine systems, which serves as a surrogate measure of capture performance. A comprehensive dataset compiled from 2969 experimental data from literature sources was curated to train and validate the models. Our results demonstrate that incorporating critical properties as inputs improves the accuracy of CO 2 solubility predictions in MDEA-based amine solutions compared to baseline models that neglect property-based descriptors. The performance comparison of the models reveals that the GBoost model delivers the best overall predictive performance, with the lowest MSE (0.078 ×10 - ⁴), MAE (0.160 ×10 - ²), and SD (0.090), along with the highest determination coefficient (R² = 0.994) on the test data. The ML/DL approaches show strong potential for rapid screening and optimization of amine blends for enhanced CO 2 capture, offering a data-driven framework that complements existing thermodynamic models. These findings contribute to a deeper understanding of the role of amine critical properties in CO 2 capture performance and support the design of more effective capture systems. • Application of machine/deep learning in predicting CO 2 capture by amine mixtures. • Using amine critical properties as input enhance CO 2 solubility prediction. • GBoost achieved MSE= 0.078 × 10⁻⁴, MAE= 0.160 × 10 - ², R²= 0.994 on test data. • Data-driven framework to screen and optimize amine blends for capture.
Sheikhshoaei et al. (2026) studied this question.