Feature selection for global tropospheric ozone prediction based on the BO-XGBoost-RFE algorithm.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Ozone is a critical air pollutant affecting human health, regional air quality, and ecosystems.
• Accurate prediction of long-term ozone concentration is essential for environmental management and public health.
• Existing methods may not fully optimize feature selection for complex environmental data.

Purpose of the Study:

• To develop and evaluate a novel feature selection model for predicting long-term average ozone concentration.
• To compare the performance of different machine learning algorithms using an optimized feature subset.
• To enhance the accuracy of global ozone concentration predictions.

Main Methods:

• Utilized geographic and environmental data from 5577 global monitoring sites (2010-2014).
• Proposed a Bayesian optimization-based XGBoost-Recursive Feature Elimination (BO-XGBoost-RFE) model for feature selection.
• Applied various machine learning algorithms (including Random Forest and XGBoost) for ozone concentration prediction on the selected optimal feature subset.

Main Results:

• The BO-XGBoost-RFE model achieved higher prediction accuracy compared to using all features or Pearson correlation-based feature selection.
• Random Forest demonstrated the highest prediction accuracy among the evaluated models.
• The XGBoost prediction model exhibited the most significant improvement in accuracy.

Conclusions:

• Bayesian optimized Recursive Feature Elimination with XGBoost is effective for selecting optimal features in ozone concentration prediction.
• Machine learning models, particularly Random Forest and XGBoost, can accurately predict long-term ozone concentrations with optimized feature sets.
• The developed methodology offers a promising approach for improving global air quality monitoring and forecasting.