Machine Learning-Based Binary Classification of Air Quality Using Pollutant Concentrations

Abstract

The primary means to protect the health of people and aid in making environmental decisions is air quality monitoring. This study creates an air quality machine learning model that analyses the concentration of pollutants to form binary air quality models. The target variable was formulated by the researchers as the categories of the original AQI_Bucket were converted into two classes: Safe (0) and Unhealthy (1). The preprocessing pipeline addressed the missing data by using numerical data processing, scaling algorithms, and converting categorical data to other forms. The authors developed three new characteristics PM_ratio, NO ratio and Gas Total that enhanced their ability to quantify the association of pollutants with each other. The researchers developed training and testing sets, using an 80:20 data split, to evaluate three machine learning algorithms, namely, Logistic Regression, Random Forest, and XGBoost. The team was able to assess the performance of the models using six measures: Accuracy, Precision, Recall, F1-score, ROC-AUC and confusion matrixes. The results demonstrated that ensemble models provided better performance results than Logistic Regression. The accuracy of Logistic Regression was 0.887 and the AUC of 0.957, whereas the accuracy of Random Forest was 0.918 and its AUC was 0.976. XGBoost produced the best overall performance with an accuracy of 0.919, precision of 0.938, recall of 0.930, F1-score of 0.934 and AUC of 0.975. The study findings are valid in showing that ensemble learning methods are effective in binary air quality classification and simplify environmental monitoring activities.