Selective adsorption processes for fructooligosaccharides separation by activated carbon and zeolites through machine learning

Abstract

Fructooligosaccharides (FOS) separation and purification are crucial for industrial applications where adsorption methods are widely used. However, some specific process characteristics can affect the selectivity of the adsorption. In this work, differences between the variables affecting the FOS separation by fixed-bed column with activated carbon and zeolite as adsorbents were identified and analyzed. Extreme Gradient Boosting (XGBoost) and Shapley Additive explanation (SHAP) methodology were used to determine the best process conditions for each adsorbent considering the most relevant variables such as temperature, time and ethanol concentration. XGBoost showed high predictive power for both adsorbents, reaching R of 0.84–0.91 for activated carbon and 0.87–0.98 for zeolite. According to SHAP, the order of importance of the variables for FOS separation for the two adsorbents was time, ethanol concentration, and temperature. Activated carbon shows selectivity for FOS at low ethanol concentrations (7.95% v/v). Zeolite required ethanol concentrations about 8 times higher than activated carbon. From SHAP it can be concluded that activated carbon is impressively efficient in improving FOS selectivity. Results showed that machine learning is a valuable tool to better understand the optimal conditions for FOS separation, allowing for higher recovery rates, efficiencies at lower ethanol concentrations, and consequently lower costs.