This study investigates the application of Machine Learning (ML) models for classifying feedwater quality in a large Reverse Osmosis (RO) desalination plant in Morocco. The main objective is to anticipate membrane fouling and improve operational efficiency, to reduce water production costs. Five classification algorithms were evaluated: eXtreme Gradient Boosting (XGB), Random Forest (RF), Multilayer Perceptron (MLP), K-Nearest Neighbors (K-NN), and Support Vector Machine (SVM). The models were assessed using key performance metrics, including accuracy, precision, recall, F1-score, and the Matthews Correlation Coefficient (MCC). The K-NN classifier achieved the highest accuracy (99.23%), followed by MLP (98.93%), XGB (98.8%), RF (98.6%), and SVM (96.56%). In terms of robustness, K-NN also attained the best MCC score (98.52%), confirming its strong predictive capabilities. The XGB and RF models followed closely, with MCC scores of 97.68% and 97.30%, respectively, highlighting their reliability in handling data variations. The MLP classifier demonstrated solid performance (MCC = 97.94%), though further hyperparameter tuning could enhance its ability to detect minority-class instances. Conversely, SVM exhibited the lowest performance (MCC = 93.40%), suggesting a higher sensitivity to data imbalances. These findings indicate that K-NN, XGB, and RF are particularly well suited for real-time monitoring of RO systems, enabling early detection of poor water quality and proactive maintenance strategies. By leveraging ML-driven predictive analytics, this research contributes to optimizing water treatment processes, extending membrane lifespan, and ultimately reducing operational costs, paving the way for more efficient and sustainable desalination operations.