Determining Key Factors Influencing Ozone Formation in Taichung City, Taiwan Using Machine Learning Models

Ozone pollution remains a significant environmental challenge in urban areas, with elevated ground-level ozone posing risks to public health, ecosystems, and climate stability. In Taichung City, Taiwan, rapid urbanization and industrial activities have contributed to deteriorating air quality, making it crucial to identify the key factors driving ozone formation for effective mitigation strategies. This study employs machine learning (ML) models, including Random Forest (RF) and eXtreme Gradient Boosting (XGBoost), to analyze ozone pollution in Taichung. Moreover, feature importance analysis is used to identify the key factors driving ozone variability, including volatile organic compounds (VOCs), nitrogen oxides (NO and NO₂), and meteorological variables such as temperature, humidity, wind speed, and solar radiation. The models were trained and tested on the hourly observational data collected from the Urban Air Pollution Research Station (UAPRS) in Taichung City from January to December 2023. To enhance the models’ accuracy, GridSearchCV is utilized to select optimal parameters and reduce the risk of overfitting. Preliminary results indicated that the number of predictors impacts ML performance—RF outperforms XGBoost when fewer predictors are used. However, with a more comprehensive set of predictors, XGBoost demonstrated superior performance, achieving determination coefficients of 0.945 and 0.886 for the training and test datasets, respectively. Feature importance analysis revealed that the top three contributors to ozone variability in 2023 were NO (44%), humidity (19%), and NO₂ (12%). For high ozone episodes, NO, humidity, and solar radiation were identified as the key drivers. By combining the predictive power of ensemble ML techniques with feature importance analysis, this study provides valuable insights into the interactions between chemical and meteorological factors driving ozone formation. The results highlight the relative significance of these factors in influencing ozone levels and provide actionable insights for air quality management in Taichung. Additionally, the study demonstrates the potential of ML models as powerful tools for advancing urban air quality research, with implications for policy interventions and future environmental studies. Future work will focus on refining the models to predict ozone episodes in real time and exploring their applicability to other rapidly urbanizing cities facing similar air quality challenges.