Improving groundwater quality predictions in semi-arid regions using ensemble learning models.

Area of Science:

• Environmental Science
• Hydrology
• Data Science

Background:

• Groundwater is a critical freshwater source in arid and semi-arid regions.
• Monitoring groundwater quality is vital for effective environmental management.
• Water quality parameters like TDS and pH are essential indicators.

Purpose of the Study:

• To compare the performance of nine ensemble and regular machine learning (ML) methods for predicting TDS and pH.
• To evaluate the effectiveness of ensemble techniques in enhancing ML model accuracy.
• To identify the most accurate models for groundwater quality prediction in semi-arid conditions.

Main Methods:

• Utilized standard ML models: Multilayer Perceptron Neural Network (MLPNN), Classification and Regression Trees (CART), and Chi-square Automatic Interaction Detection (CHAID).
• Developed ensemble versions of these models using Bagging (BG) and Boosting (BT) techniques.
• Assessed model performance using Standard Root Mean Square Error (SRMSE) for TDS and pH prediction.

Main Results:

• Standard ML models showed comparable results for TDS prediction, with MLPNN being the most accurate.
• Predicting pH was more challenging for all models.
• Ensemble techniques, on average, improved regular model accuracy by 22.68%, with boosting outperforming bagging.
• CHAID-BT and CHAID-BG models achieved the highest accuracy for TDS and pH prediction, respectively.

Conclusions:

• Ensemble machine learning techniques offer a highly accurate alternative for predicting groundwater quality parameters.
• These methods are valuable tools for sustainable water resource management in semi-arid regions, addressing challenges like water scarcity and pollution.
• The study highlights the significant potential of ensemble ML in improving water quality monitoring and management strategies.