ABSTRACT

Air pollution has become a critical environmental concern in India, posing severe health risks to millions of people. This project aims to predict the air quality levels—categorized as Good, Moderate, or Poor—using machine learning algorithms applied to historical air pollution data. The dataset, sourced from Kaggle, comprises air quality measurements collected from various Indian cities between 2015 and 2020. Key pollutant indicators such as PM2.5, PM10, NO, NO2, NOx, NH3, CO, SO2, O3, Benzene, and Toluene are used as input features, while the AQI Bucket serves as the target label. The dataset underwent thorough preprocessing including missing value imputation, categorical encoding, and class balancing using SVMSMOTE and undersampling techniques. Six classification models were implemented: Logistic Regression, MLP Classifier, Decision Tree, Random Forest, AdaBoost, and XGBoost. A custom prediction class was developed for real-time input processing and prediction. This system demonstrates how machine learning can effectively assist in environmental monitoring, enabling authorities and citizens to take timely actions to mitigate the impact of poor air quality.