Abstract

Air pollution has emerged as a major environmental and public health concern, particularly in rapidly urbanizing countries such as India. Accurate prediction of air quality levels is essential for timely decision-making and effective pollution control strategies. This paper presents a machine learning–based air quality prediction system that classifies air quality into three categories: Good, Moderate, and Poor using historical air pollution data. The proposed approach utilizes pollutant concentration data collected from multiple Indian cities between 2015 and 2020, including key indicators such as PM2.5, PM10, NO, NO₂, NOx, NH₃, CO, SO₂, O₃, Benzene, and Toluene. To address the inherent class imbalance in air quality datasets, a hybrid resampling strategy combining Synthetic Minority Oversampling Technique based on Support Vector Machines (SVMSMOTE) and random undersampling is employed. Several supervised machine learning models, including Logistic Regression, Decision Tree, Random Forest, Multi- Layer Perceptron, AdaBoost, and XGBoost, are trained and evaluated using standard performance metrics. Experimental results demonstrate that ensemble-based models, particularly XGBoost, achieve superior classification performance, attaining an accuracy of 94.25% with balanced precision and recall across all classes. The proposed system offers an efficient, scalable, and data-driven solution for air quality prediction and can be extended to support real-time environmental monitoring and public health decision-making.