A Novel Air Quality Labelling and Classification Approach Using Supervised Learning

Abstract: Air quality assessment is crucial for environmental monitoring, public health and decision-making. Air quality does not depend solely on the concentration of certain gasses; rather, it is also influenced by other pollutants that are challenging to measure individually for further investigations. In this study, we used the UCI Air Quality dataset, which included key pollutants such as CO, NO₂, NOx, benzene, temperature, humidity and sensor data for other forms of pollutants. We proposed a novel labeling scheme based on weighted pollutant concentrations, enabling more precise air quality classification into Good, Moderate, and Unhealthy categories. After that, we evaluated five supervised learning models—Random Forest, Decision Tree, Support Vector Machine, K-Nearest Neighbors, and Gradient Boosting—for classification, considering all types of measured pollutants, and assessed their performance using accuracy, confusion matrices, classification reports, and ROC-AUC curves. Our research also highlights the potential of AI-driven techniques in comprehensive air quality assessment as well as real-time air pollution prediction and classification for environmental protection.