COMPARATIVE ANALYSIS OF MACHINE LEARNING MODELS FOR AIR QUALITY FORECASTING IN MAHARASHTRA

Abstract

This paper addresses the machine learning (ML) based prediction of air pollution in various cities of the state Maharashtra. To conduct the study data collected by the Maharashtra Air Quality Monitoring Network, were used to overcome data scarcity and train the models. A comparison of PM2.5, PM10, nitrogen dioxide (NO2), sulfur dioxide (SO2), and ozone (O3) over the whole year of 2023 provides a clear picture of how the air quality index (AQI) changes over time. The most accurate models were the extreme gradient boosting (XGBoost) model (98.83%) and the naive Bayes classifier (NBC) (95.58%). The support vector machine (SVM) reported the lowest accuracy of 81%, which is still improving. This research study, in addition to illustrating how ML-based models can help predict future air quality trends, also highlights the advantages of using synthetic data to increase the accuracy of the predicted outputs. These results provide a useful perspective for policymakers in the formulation of effective interventions that can be implemented to alleviate the air pollution problem and ensure that the urban centers in Maharashtra are transformed into places where people can live quality lives. This article establishes the relevance of further research on the use of synthetic data and machine learning in the fight against poor air quality.