Optimization of adsorption capacity for CO2/N2 mixture via RSM and machine learning

• A comprehensive review of the effective parameters in the adsorption process is conducted, • Input variables and performance criteria for adsorption capacity calculation are reported. • Adsorption capacity of CO₂ and N₂ on various solid adsorbents investigated using computational methods, • Molecular polarizability and FTIR peak characteristics are considered as input for computational methods for the first time, • High precision for all computational methods investigated, This study models the equilibrium adsorption capacity of CO₂ and N₂ on various solid adsorbents using response surface methodology (RSM) and machine learning (ML) approaches, including multilayer perceptron (MLP), radial basis function (RBF), and support vector machine (SVM) networks. The input parameters included textural properties (specific surface area, pore volume, and nominal pore diameter), process conditions (temperature and pressure), and adsorbate characteristics such as molecular polarizability and FTIR peak position. The models were developed and optimized to predict adsorption capacity under equilibrium conditions. Among the tested models, the RBF network achieved the highest accuracy (R² = 0.9999), followed by MLP (R² = 0.9996) and SVM (R² = 0.9980), while the RSM model showed slightly lower precision (R² = 0.9884). Sensitivity analysis identified polarizability as the most significant positive factor and FTIR peak intensity as the most influential negative contributor, emphasizing the key role of electrostatic interactions and surface functionality in gas–solid systems. The integration of surface chemistry descriptors and molecular properties into ML-based frameworks provides a powerful and efficient strategy for predicting adsorption behavior and guiding the design of advanced adsorbents for CO₂ capture and separation processes.