Prediction of Soil Pollution Risk Based on Machine Learning and SHAP Interpretable Models in the Nansi Lake, China

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Nansi Lake faces significant soil pollution challenges, primarily from heavy metals.
• Assessing and predicting soil pollution is crucial for environmental protection and risk management.

Purpose of the Study:

• To develop and validate a machine learning model for predicting soil pollution risk in the Nansi Lake region.
• To integrate traditional indices like Pollution Load Index (PLI) and Potential Ecological Risk Index (PERI) with machine learning for enhanced accuracy.
• To identify key heavy metal pollutants contributing to soil contamination.

Main Methods:

• Employed machine learning models including Support Vector Machine (SVM), Decision Tree Classifier (DT), Random Forest (RF), and XGBoost.
• Utilized statistical characteristics, PLI, and PERI to create a three-class pollution risk categorization (no, low, high risk).
• Applied SHapley Additive exPlanations (SHAP) for model interpretability to identify significant pollution drivers.

Main Results:

• XGBoost model achieved the highest prediction accuracy at 93% for soil pollution risk.
• SHAP analysis identified cadmium (Cd) and mercury (Hg) as significant contributors to soil pollution risk.
• The integrated method successfully categorized soil pollution into 'Class0-no risk', 'Class1-low risk', and 'Class2-high risk'.

Conclusions:

• Machine learning, particularly XGBoost, offers a highly accurate method for predicting soil pollution risk.
• Cadmium and mercury are critical pollutants requiring targeted management strategies in the Nansi Lake region.
• The developed assessment framework provides a robust tool for environmental monitoring and decision-making.