Forcasting Water Contamination in Rural Zones Using Machine Learning

Background

Waterborne disease remains a major public health risk in rural areas where access to treated water is limited and testing is infrequent.

What We Built

We built a predictive model using historical rainfall, elevation, and well-testing data to forecast when and where microbial contamination is most likely to spike in groundwater supplies.

Modeling & Research

The model was trained using SVM and random forest classifiers, labeling high-risk days based on past contamination outbreaks. Input features included cumulative rainfall, proximity to livestock, and surface runoff potential.

Impact

With minimal inputs, the model identifies high-risk time windows and enables preventative interventions such as chlorination alerts or mobile testing dispatch.

Outcome

Accepted for internal publication under NSRI’s Environmental Research Program (2025).