Computational evaluation using machine learning for analysis of membrane desalination process powered by solar energy.

Numerical simulation of mass transfer in membrane contactors for water purification was carried out in this study. The models were developed to estimate permeate volume, a critical performance variable in membrane desalination. We analyzed and compared the accuracy of models: Multilayer Perceptron (MLP), Extreme Gradient Boosting (XGBoost), and a Stacking ensemble method combining MLP and XGBoost. Data preprocessing techniques were applied to improve model performance, including outlier detection using Z-scores and log transformation for normalization. A distinctive contribution of this work is the innovative application of the Jellyfish Optimizer (JO) for hyperparameter tuning, which enhanced model robustness and set it apart from conventional optimization approaches. Unlike conventional optimization algorithms, JO's biologically inspired dynamics offers a novel balance between exploration and exploitation, leading to superior tuning of model hyperparameters. The models' predictive accuracy was evaluated using R², RMSE, and MAE with an 85%-15% train-test split. The Stacking model outperformed single models, achieving R² values of 0.99060 (training), 0.97627 (cross-validation), and 0.99043 (testing), along with the lowest test errors (RMSE = 6.17921), confirming its efficiency in predicting permeate volume. The results show that stacking significantly improves prediction accuracy in complex tasks like membrane desalination.