Smart Waters: Harnessing Machine Learning to Predict Water Quality in a Tropical Andean Watershed

This study is based on the application of machine learning to compare regression tree models based on data with a reduced set of parameters using the National Sanitation Foundation Index (WQI NSF) to estimate water quality in the Yanuncay River watershed. Physical, chemical, and biological parameter data were collected from the watershed and equations derived from curve fits were used to calculate the WQI NSF. The results showed that the regression random forest model trained with three parameters: fecal coliforms, pH and nitrates, was the most suitable option. This model demonstrated consistent performance, with an R² of 0.930 and a standard deviation of 0.026. The importance of fecal coliforms and nitrates as key indicators of contamination were highlighted, and pH was considered crucial due to its ease of sampling in the field and low requirement of specialized equipment. Thus, this study highlights the importance of continuous and long-term water quality monitoring in the Yanuncay River watershed and suggests that regression tree-based models can optimize monitoring requirements without compromising accuracy in estimating the WQI NSF.