Developing a framework for classifying water lead levels at private drinking water systems: A Bayesian Belief Network...

Area of Science:

• Environmental Science
• Public Health
• Data Science

Background:

• Lead in drinking water is a significant public health concern, causing neurological damage even at low exposure levels.
• Private water systems (wells, springs) lack the regulatory oversight of public systems, potentially increasing user risk.
• Predictive modeling is needed to assess and mitigate lead exposure risks in private water sources.

Purpose of the Study:

• To explore and develop modeling approaches for predicting lead contamination risk in private drinking water systems.
• To investigate the interactions between household characteristics, geological factors, and water quality parameters in relation to lead levels.
• To establish a knowledge discovery framework integrating data preprocessing and Bayesian classification methods.

Main Methods:

• Utilized Bayesian Network approaches, including Naive Bayes and Tree-Augmented Naive Bayes, to model lead risk.
• Employed feature selection techniques (forward/backward selection) and various data discretization methods.
• Applied Bayesian inference to fit conditional probability tables for identified Directed Acyclic Graphs (DAGs).
• Analyzed data from the Virginia Household Water Quality Program (VAHWQP) involving 2,146 households.

Main Results:

• Naive Bayes classifiers demonstrated superior performance in predicting lead levels, based on recall and precision.
• Copper concentration emerged as the most significant predictor of lead, followed by county, pH, and on-site water treatment.
• Discretization methods significantly impacted model performance, while feature selection had a marginal effect.

Conclusions:

• The developed Bayesian framework effectively models lead risk in private water systems.
• Copper, pH, and specific household/location factors are critical indicators for elevated lead levels.
• Findings can inform targeted lead testing and treatment strategies for vulnerable private well users.