Optimizing coastal groundwater quality predictions: A novel data mining framework with cross-validation,...

Area of Science:

• Environmental Science
• Hydrogeology
• Data Mining

Background:

• Groundwater quality is critical for resource management, particularly in contaminated coastal regions like Bangladesh.
• Existing data mining algorithms (DMAs) require optimization for accurate groundwater quality modeling.

Purpose of the Study:

• To investigate the applicability of DMAs, including Gaussian Process Regression (GPR), Bayesian Ridge Regression (BRR), and Artificial Neural Network (ANN), for predicting coastal groundwater quality.
• To enhance model accuracy using optuna-based hyperparameter optimization and combined cross-validation (CV) and bootstrapping (B) methods.
• To identify spatial groundwater quality patterns and key influencing variables.

Main Methods:

• Developed and compared six predictive models: optuna-GPR, optuna-BRR, ANN (CV), ANN (B), and benchmark models.
• Utilized self-organizing map (SOM), spatial autocorrelation, and fuzzy logic to analyze 12 physicochemical variables from 67 wells.
• Calculated the entropy-based coastal groundwater quality index (ECWQI) and normalized it (ECWQIn).

Main Results:

• Artificial Neural Network (ANN) models, both ANN (CV) and ANN (B), demonstrated superior performance with low RMSE and high R² and CC values.
• Sulfate (SO₄²⁻), chloride (Cl⁻), and fluoride (F⁻) were identified as significant predictors of groundwater quality.
• SOM analysis revealed four distinct spatial patterns, with F⁻ and SO₄²⁻ exhibiting high spatial autocorrelation, impacting groundwater quality.

Conclusions:

• The ANN model is highly effective for predicting groundwater quality in coastal environments.
• Optimized DMAs, particularly ANN, provide a reliable tool for real-time monitoring and sustainable groundwater resource management.
• Understanding spatial patterns and key chemical variables is crucial for effective water management strategies.