Enhancing nitrate and strontium concentration prediction in groundwater by using new data mining algorithm.

Area of Science:

• Environmental Science
• Hydrogeology
• Data Mining

Background:

• Groundwater is vital for domestic use and agriculture, especially in the Campanian Plain, Italy.
• Ensuring the sustainability of groundwater resources is critical for the region's future.
• Groundwater quality is threatened by contaminants such as nitrate and potentially increasing elements like strontium.

Purpose of the Study:

• To apply novel data mining algorithms for predicting nitrate and strontium concentrations in groundwater.
• To compare the performance of Gaussian Process (GP) with other machine learning models (M5P, Random Forest, Random Tree).
• To identify optimal input variables for robust groundwater quality modeling.

Main Methods:

• Utilized a groundwater quality database of 246 samples from the Campanian Plain.
• Employed 10-fold cross-validation for model training and testing.
• Applied Gaussian Process (GP), M5P, Random Forest (RF), and Random Tree (RT) algorithms.
• Selected input variables based on correlation coefficients and evaluated model performance using quantitative criteria.

Main Results:

• Gaussian Process (GP) demonstrated superior performance in predicting both nitrate and strontium concentrations compared to RF, M5P, and RT.
• Variables with low correlation coefficients, alongside highly correlated ones, are important for reliable model predictions.
• Model structure and data characteristics significantly impact prediction accuracy.

Conclusions:

• Gaussian Process (GP) is a highly effective tool for predicting groundwater contaminant concentrations.
• Data mining offers a robust approach for assessing and managing groundwater quality.
• Consideration of diverse input variables enhances the reliability of hydrogeological models for sustainable water resource management.