Modeling Surface Water Quality Using K-Nearest Neighbors and Random Forest

Surface water quality monitoring is crucial for effective natural resource management. This study applies a machine learning approach to classify pollution levels of surface water bodies in Costa Rica. Two algorithms, K-Nearest Neighbors (KNN) and Random Forest (RF), were employed to process data from the National Plan for Monitoring the Quality of Surface Water Bodies (2015-2020). The classification was based on the Dutch Index for water quality assessment, ranging from "no contamination" to “severe contamination”. The relative weight of Biological Oxygen Demand (BOD), Oxygen Saturation (OS), Ammoniacal Nitrogen (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathrm{N}-\text{NH}_{4}{ }^{+}$</tex>), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{p H}$</tex>, Temperature (T), and Total Suspended Solids (TSS) was evaluated. The best performance for both algorithms was achieved using only the most impactful \includegraphics[max width=\textwidth]{10885395-1} KNN and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 3. 4 \%}$</tex> for RF. These results suggest that monitoring these select variables can reduce the cost of water quality management, in countries like Costa Rica, with no “very severe” levels of water pollution.