Machine learning-based prediction of CO2 solubility in methyldiethanolamine solutions: A comparative study

• Developed machine learning models to predict CO₂ solubility in MDEA-based solutions. • Compiled an extensive dataset of 2969 experimental data points from the literature. • Evaluated three soft computing techniques: RBF-NN, GPR, and MLP-NN. • Gaussian Process Regression (GPR) achieved the highest accuracy (AARE = 4.94%, R² = 97.5%). • Sensitivity analysis identified key factors influencing CO₂ solubility. • Provides a reliable predictive tool for carbon capture applications. Methyldiethanolamine (MDEA) is a widely used solvent in carbon capture processes owing to its high absorption capacity. However, there is a lack of comprehensive predictive tools for estimating CO 2 solubility in MDEA-based solution. To fulfil this research gap, in the current study, 2969 experimental data pertinent to the CO 2 dissolution in MDEA solutions blended with various co-solvent, including water, amines, ionic liquids, electrolytes, etc., have been collected from the literature. The foregoing databank envelop a widespread range of pressures and temperatures. In order to construct robust models, three heuristic soft computing methods, including radial basis function neural network (RBF-NN), gaussian process regression (GPR) and multilayer perceptron neural network (MLP-NN) were employed. Despite the satisfactory performance of all intelligent models, the one designed based on the GPR method gave the superior accuracy with average absolute relative error (AARE) and R 2 values of 4.94% and 97.5%, respectively, for the testing dataset. Moreover, it estimated more than 91% of the analyzed samples within ± 15 error margin. A statistical investigation through the William’s plot implied the fact that both the collected databank and the suggested predictive tools benefit from high credibility. The novel models also favorably described the absorption capacity of diverse MDEA-based solutions under a wide range of operating conditions. Finally, the order of significance of influential factors in controlling solubility was determined based on a sensitivity analysis.