Solar stills present a sustainable and eco-friendly solution to produce freshwater in arid areas. However, complicated heat transfer mechanisms and the computational demands of conventional simulation techniques frequently limit their effectiveness. In order to accurately and effectively predict the water temperature and production of solar stills, this study suggests a machine learning-based approach, allowing for real-time performance assessment and design improvement. Utilizing a compact dataset with 236 samples produced by Computational Fluid Dynamics (CFD) simulations, a Random Forest Regressor (RFR) is trained. In order to improve the interpretability of the model's predictions, feature importance is analyzed using SHapley Additive exPlanations (SHAP). Key parameters like solar radiation, ambient temperature, wind speed, fin height, number of fins, and Phase Change Material (PCM) type were varied to create the dataset. Hyperparameter tuning was used to train and optimize the RFR model in order to increase its predictive performance. With a combined training and prediction time of roughly two seconds, the optimized RFR model shows its ability to precisely predict the solar still's performance in real time with a Mean Absolute Percentage Errors (MAPE) of 7.34 % for productivity and 2.46 % for water temperature, as evaluated on the unseen testing dataset. According to SHAP data, the most important variables influencing the performance of solar stills were fin height, wind speed, ambient temperature, and the application of lauric acid PCM. These results highlight the model's excellent accuracy and efficiency, providing a useful and less computationally demanding substitute for conventional CFD simulations in solar still analysis.