• Explored the correlation between membrane structure parameters and operating conditions with water/salts selectivity and reveal the importance of the diverse features of four typed salts

• Random Forest and Gradient Boost models were employed to learn from relevant variables and assess their importance

• Features importance in the different pore size ranges was emphatically evaluated

• Comprehensive insights can be further leveraged to tackle performance modulation and oriented design of multi scenarios NF membranes.