Water pollution control is a crucial aspect of environmental safety and sustainable development. Public Private Partnerships (PPP) play a significant role in this control but are exposed to several risks. This study proposes a new risk classification prediction model for water treatment PPP projects to address these risks more effectively than traditional methods. The proposed model includes four key areas of risk: natural environment, ecological environment, socio-economic, and engineering entity. The study examines the correlation between these risk factors and project risk levels and develops an ensemble learning model based on Stacking for risk prediction. This model improves performance by using a weighted voting mechanism to adjust the importance of base learners. This model was tested using data from Phase I of the Jiujiang City water environment system project, demonstrating its effectiveness and accuracy. The proposed model outperforms other traditional machine learning models in terms of accuracy, macro-average precision, recall, and F1-score. Thus, it provides an effective method for risk classification prediction in water treatment PPP projects.