Novel hybrid approach of random forest and stacking ensemble to improve fresh water yield prediction in mobile wick solar still

Freshwater scarcity remains a global concern, and solar stills offer a sustainable solution, but they have low yields. This work employs machine learning to predict and optimize solar still productivity from environmental and operational data. Random Forest, AdaBoost, XGBoost, and Linear Regression were trained (80:20 train:test split) with GridSearchCV for hyperparameter tuning. Random Forest achieved the highest accuracy (train R² = 0.975, test R² = 0.830) and surpassed both XGBoost (test R² = 0.784) and the stacking ensemble (test R² = 0.783). Feature importance analysis identified solar still interior temperature and global irradiation as the most influential parameters. These findings highlight the potential of machine learning to enhance solar still performance. Future work will focus on integrating real-time environmental data and expanding the model deployment for adaptive freshwater production techniques.