Efficient drinking water management requires anticipating both the availability of the resource and the evolution of the parameters that determine its quality, especially in contexts of increasing climate variability. In this scenario, the ability to project flow rates and relevant contaminants is a key tool to support operational decision-making in drinking water treatment systems. This work develops a predictive methodology based on machine learning for estimating and projecting flow rates and the water quality parameters of total dissolved solids (TDS), nitrates (NO−3), and turbidity, using historical data and external climatic and hydrological variables. Specifically, the analysis considers available monthly pollutant records for the period January 2021 to May 2025, a historical flow series corresponding to the period January 2000 to May 2025, and external climatic variables (temperature, precipitation, and accumulated snow) that include historical data from January 2000 and future projections up to June 2030. The study is being conducted at the San Juan Drinking Water Treatment Plant (DWTP), located in the Llolleo sector of the San Antonio commune, Valparaíso Region, and is being carried out in collaboration with the water utility Esval, as part of an applied study of a real-world operating system. The proposed methodology is based on the implementation of the Random Forest model, which allows for capturing nonlinear relationships and complex temporal dependencies among the analyzed variables. For each variable of interest, different model configurations are evaluated, considering the selection of the optimal number of predictor variables and the number of trees in the forest, using absolute and relative error metrics. Furthermore, the relative importance of the selected variables is analyzed, projections are generated for 1- and 5-year horizons, and the predictive performance of the model is validated by comparing projected and observed values. The results obtained show that the Random Forest model is capable of adequately representing the temporal dynamics of the flow rate and the analyzed contaminants, providing projections consistent with the historical behavior of the system and acceptable error levels during the validation periods. Overall, this work provides a replicable methodological basis that can be extended to other treatment plants and used to support the preventive and operational management of drinking water quality.