Improving the robustness of beach water quality modeling using an ensemble machine learning approach

Area of Science:

• Environmental Science
• Data Science
• Microbiology

Background:

• Microbial pollution in beach water poses health risks to swimmers.
• Traditional culture-based water quality assessments have limitations.
• Existing machine learning models show variable performance across different locations and time.

Purpose of the Study:

• To introduce and evaluate a novel ensemble machine learning approach, model stacking, for predicting beach water quality.
• To compare the performance of model stacking against five individual machine learning models.

Main Methods:

• Developed a two-layered ensemble model stacking approach.
• Utilized five individual machine learning models: multiple linear regression, partial least square, sparse partial least square, random forest, and Bayesian network.
• Applied the model stacking approach to water quality data from three beaches along eastern Lake Erie, New York.

Main Results:

• The model stacking approach consistently achieved high prediction accuracy.
• The stacking model ranked 1st or 2nd in accuracy each year across the studied beaches.
• Yearly average accuracies for the stacking model were 78%, 81%, and 82.3% at the three beaches.

Conclusions:

• Model stacking offers a reliable and effective method for predicting beach water quality.
• This ensemble approach addresses the variability issues of individual machine learning models.
• The model stacking approach shows promise for solving other environmental prediction challenges.