Comparative study of machine learning models for evaluating groundwater vulnerability to nitrate contamination.

Area of Science:

• Environmental science
• Hydrology
• Machine learning applications

Background:

• Effective groundwater contamination vulnerability assessment is vital for watershed management and pollution prevention.
• Machine learning offers advanced tools for complex environmental modeling tasks.

Purpose of the Study:

• To compare the performance of Radial Basis Neural Networks (RBNN), Support Vector Regression (SVR), and ensemble Random Forest Regression (RFR) for groundwater contamination vulnerability evaluation.
• To identify the most accurate machine learning model for predicting vulnerable areas.

Main Methods:

• Utilized eight vulnerability factors from the modified DRASTIC model (MDM) as input data.
• Employed adjusted vulnerability index (AVI) with nitrate values as the output for modeling.
• Evaluated model performance using statistical criteria: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), coefficient of determination (r²), and Receiver Operating Characteristic/Area Under the Curve (ROC/AUC).

Main Results:

• The ensemble Random Forest Regression (RFR) model demonstrated superior performance compared to standalone SVR and RBNN models.
• Ensemble RFR exhibited flexibility and robustness, retaining optimal solutions throughout the performance evaluation.
• The vulnerability map generated by RFR provided more accurate predictions of areas susceptible to contamination.

Conclusions:

• Ensemble RFR is a robust and effective tool for improving groundwater contamination vulnerability assessments.
• The findings contribute to enhanced environmental safety by enabling better prediction and management of groundwater contamination risks.