Interpreting optimised data-driven solution with explainable artificial intelligence (XAI) for water quality...

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Water pollution and scarcity in Saudi Arabia threaten sustainable development goals.
• Accurate water quality assessment is crucial for effective water management decisions.

Purpose of the Study:

• To develop optimized data-driven models for comprehensive water quality assessment.
• To enable informed decisions for sustainable water resource management in Saudi Arabia.

Main Methods:

• Calculated Water Quality Index (WQI) using entropy-weighted arithmetic and World Health Organization (WHO) standards.
• Employed machine learning (ML) models (Decision Trees, Random Forests (RF), Gradient Boosting Machines (GBM), Deep Neural Networks (DNN)) via the H2O API.
• Utilized explainable artificial intelligence (XAI) for model interpretation, including feature importance and partial dependence plots.

Main Results:

• Over 35% of water samples were classified as 'unsuitable' for consumption.
• Fluoride and residual chlorine showed the highest and lowest entropy weights, respectively.
• Optimized RF (model 79) and DNN (model 81) achieved high predictive accuracy (R²=0.96 and 0.97).
• Key parameters like nitrate, total hardness, and pH significantly influenced WQI predictions.

Conclusions:

• Data-driven models, particularly RF and DNN, provide robust water quality assessment.
• XAI methods reveal critical water quality parameters for targeted improvement strategies.
• The study demonstrates the potential of advanced analytics for sustainable water resource management in Saudi Arabia.