A Comparative Analysis of Explainable Artificial Intelligence Models for Electric Field Strength Prediction over...

Area of Science:

• Environmental Science
• Computer Science
• Public Health

Background:

• Wireless device proliferation raises concerns about electromagnetic field (EMF) exposure.
• Accurate mapping of EMF is crucial for environmental monitoring and public health assessments.

Purpose of the Study:

• To identify the optimal machine learning model for creating detailed electric field strength maps in urban areas.
• To leverage sensor data and explainable AI for enhanced understanding of EMF distribution.

Main Methods:

• Collected a novel dataset from sensors measuring EMF, population density, urbanization, and building characteristics.
• Analyzed 566 machine learning models (k-NN, XGBoost, Random Forest, Neural Networks, Decision Trees, Linear Regression) across eight French cities.
• Utilized SHAP analysis to determine feature importance and model interpretability.

Main Results:

• Ensemble methods, specifically Random Forests and XGBoost, demonstrated superior predictive performance over individual models.
• Machine learning models significantly outperformed the linear regression baseline.
• Feature importance varied notably between tree-based models and other approaches like k-NN and neural networks.

Conclusions:

• Machine learning, particularly ensemble methods, offers a powerful tool for dynamic electromagnetic pollution mapping.
• Sensor-driven data and explainable AI enhance the accuracy and interpretability of EMF exposure assessments.
• Findings provide a foundation for tracking EMF trends and evaluating mitigation strategies in urban environments.