How do data-mining models consider arsenic contamination in sediments and variables importance?

Area of Science:

• Environmental Science
• Geochemistry
• Data Mining

Background:

• Arsenic (As) poses a global health risk, with over 100 million people exposed.
• Drinking water standards set by WHO and Iran limit As to 10 μg/L.
• Elevated As concentrations are prevalent in several Iranian regions, necessitating risk assessment.

Purpose of the Study:

• To evaluate arsenic (As) contamination susceptibility in the Tajan River watershed, Iran.
• To compare the effectiveness of various data-mining models in predicting As contamination.
• To identify key environmental factors contributing to As contamination.

Main Methods:

• Employed six data-mining models: MARS, FDA, SVM, GLM, MDA, and GBM.
• Utilized 12 environmental factors including land use, lithology, and topography for susceptibility mapping.
• Conducted model training (70%) and validation (30%) using ROC curve analysis.

Main Results:

• GBM, MARS, and SVM identified the largest areas of high As contamination risk (30.0%, 28.7%).
• FDA, GLM, MARS, and MDA indicated significant areas of low risk (3.3%–72.0%).
• MARS, SVM, and MDA demonstrated high predictive accuracy (AUC ROC 78.9%–84.6%).
• Land use, lithology, erosion, and elevation were the most influential predictors of As contamination.

Conclusions:

• Data-mining techniques are effective, cost-efficient tools for identifying As-susceptible areas.
• Findings can guide environmental managers in mitigating As contamination in sediments and the food chain.
• The study highlights critical factors influencing arsenic contamination in the Tajan River watershed.