Classification of Air Pollution Risk Levels Using a Soft Voting Ensemble Model Based on Real-World Air Quality Monitoring Data

Abstract

Air pollution is one of the major environmental and public health challenges in rapidly urbanizing regions around the world. Elevated concentrations of airborne pollutants such as particulate matter and gaseous emissions can threaten both human health and environmental sustainability. Thus, accurate recognition of pollution risk degrees is vital for developing environmental monitoring systems and supporting decision-making in urban environmental management. We present an ensemble machine learning-based framework to classify the levels of air pollution risk using environmental, meteorological and temporal indicators extracted from real-world air quality monitoring data collected through different urban locations. Our dataset consists of major atmospheric pollutants as well as meteorological variables capturing the air pollution processes in urban areas over varying seasons. Initially evaluated several machine learning algorithms like Random Forest, Extra Trees, Support Vector Machine, Logistic Regression and Extreme Gradient Boosting. A Soft Voting ensemble model was then designed to combine the prediction strengths of all best-performing classifiers. The proposed model attained an accuracy of about 82.7% with the weighted F1-score being 0.828, thus performing better than any single models. Cross-validation validated the framework's robustness and stability, allowing analysis of feature importance to highlight PM2. 5 = the most important determinant of pollution risk levels. The findings highlight the utility of ensemble machine learning methodologies for environmental monitoring, providing greater insight into pollution exposure and informing data-driven decision making to promote sustainable air quality management