Exploration of the reverse osmosis desalination process by explainable machine learning to support sustainable development goal 6: Clean water and sanitation

Sustainable Development Goal (SDG) 6 calls for safe water, sanitation, and hygiene to safeguard health. Aligned with these goals, this study explores the reverse osmosis desalination process through machine learning, aiming to enhance the production of drinking water from brackish water and seawater in an energy efficient way. A database of 838 experimental entries from 29 studies was compiled, covering three targets: water recovery, specific energy consumption (SEC), and permeate salinity. Neural-network regressors achieved high predictive accuracy, providing a strong basis for interpretability for both water types. For example, for seawater, SHapley Additive exPlanations (SHAP) analysis revealed that pretreatment and energy recovery devices (ERDs) reduced SEC while lower flowrate and multi-stage configurations improved recovery. Decision tree models further identified practical operational routes, achieving testing accuracies of 85-93 %. For example, high seawater recovery was reliably obtained with pressures above 47 bar, feed salinity below 40,000 ppm, and multi-stage membrane configurations. On the other hand, low SEC required an ERD, feed salinity <43,400 ppm, and pH 7.9. Overall, the results indicate that machine learning is a powerful tool for detecting performance trends predictive drivers in reverse osmosis desalination, thereby supporting research and practice in sustainable freshwater production.