Predictive Model for Ammonia Absorption Capacity of Deep Eutectic Solvents Using Rough Set Theory

Abstract The direct emission of ammonia (NH 3 ) into the atmosphere from process waste gas streams brings environmental and health risks. This can be mitigated through absorption using solvents. Deep eutectic solvents (DES) were introduced as green absorption solvents with advantages including wide liquid range, high thermal stability and simple synthesis. Experimental studies indicate that NH₃ absorption capacity improves with acidic DES, higher hydrogen bond donor (HBD) to hydrogen bond acceptor (HBA) ratio, controlled water content, lower temperature and higher pressure. However, the exhaustive number of possible DES combinations formed by varying HBAs, HBDs, molar ratio and water content, makes it difficult to evaluate each combination experimentally. This combinatorial complexity together with the influence of process conditions, makes determining the optimal ranges of these variables through trial-and-error experiments resource-intensive and time-consuming. While predictive modelling offers a solution, most existing models are black-box approaches that lack interpretability, providing limited insights into how each variable affects NH 3 absorption performance. This limits their application in guiding solvent and process design. Therefore, this work developed an interpretable predictive model for NH 3 absorption in DES using rough set machine learning (RSML). RSML applies rough set theory (RST) to generate interpretable results by analysing indiscernibility relations within large datasets. Data on DES properties and process conditions affecting NH 3 absorption capacity were collected for model development. Decision rules with high certainty (> 80%) were analysed. Results revealed that acidic HBDs (i.e. phenols), lower temperatures (< 338 K), higher pressures (> 101.3 kPa) and controlled water content (0.025–0.075 wt%) enhance NH 3 absorption capacity.