A new method for drinking water quality risk assessment based on data-driven

Risk assessment of water quality plays a crucial role in sustainable management of water resource. However, evaluating drinking water quality risk for different types of water within the same framework is a challenging task. The Water Quality Index (WQI) has proven to be a cost-effective framework for assessing drinking water quality. But the conventional WQI approach is unable to deal with subjectivity, uncertainty, and boundary ambiguity involved in the assessment. To overcome these limitations, a total process improvement integrating machine learning, comprehensive weighting, and fuzzy mathematics with WQI-fuzzy and data-driven WQI (FDWQI) is proposed in this study for assessing drinking water quality. Based on the principle of index screening based on pollution risk and volatility, different index data sets were selected for different water bodies. The high area of the curve (AUC) and precision indicate that the model has been very successful and can be well applied to different types of water. The trapezoidal membership functions classified the model input parameters into desirable, fine and bad. The comparative assessment of the WQI models showed that the FDWQI predictions of the three drinking water qualities were more accurate and reasonable, and had greater interpretability. The assessment results indicate that some surface and groundwaters in the study area (73% surface water; 7% shallow groundwater; and 21% deep groundwater) have high water quality risks, with surface water having extremely severe water quality risks that are not potable. This study provides a good example of how to assess and compare the water quality risks of different water bodies under uniform criteria.