Predicting sustainable arsenic mitigation using machine learning techniques.

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Arsenic contamination poses significant global health risks.
• Sustainable mitigation strategies require accurate predictive modeling.
• Machine learning offers potential for predicting effective mitigation preferences.

Purpose of the Study:

• To evaluate and compare the performance of various machine learning models.
• To identify the most accurate and robust model for predicting sustainable arsenic mitigation preferences.
• To assess model performance based on Area Under the Curve (AUC) and Kappa scores.

Main Methods:

• Evaluated state-of-the-art machine learning classifiers including Naïve Bayes (NB), Support Vector Classification, K-Neighbors, ensemble methods (Random Forest, Decision Tree), and linear classifiers.
• Utilized Receiver Operating Characteristic (ROC) curve analysis to calculate AUC.
• Assessed model robustness using Kappa scores on training and test datasets.

Main Results:

• Gaussian Naïve Bayes (NB) achieved the highest AUC (0.82), demonstrating superior predictive accuracy.
• Ensemble classifiers like Random Forest (0.77) and Decision Tree (0.77) also showed strong performance.
• Linear classifiers generally underperformed, with the perceptron model showing the lowest AUC (0.57).

Conclusions:

• Nonlinear and ensemble machine learning classifiers are best suited for modeling complex socio-environmental data in arsenic mitigation.
• Gaussian NB is the optimal choice for predicting sustainable arsenic mitigation, especially with limited data.
• Accurate predictive models are crucial for developing effective and robust arsenic mitigation strategies.