PM 2.5 Prediction and Remark Classification

Air pollution has become a major issue in India. Fine particulate matter ($\mathbf{P M}_{2.5}$) consists of airborne particles smaller than 2.5 micrometers that can reach deep into the human respiratory system, making them one of the most harmful components of urban air pollution. This research develops predictive models using advanced ensemble learning methodsXGBoost, LightGBM, and CatBoost-along with a voting ensemble, leveraging multipollutant and temporal data from$\mathbf{1 0}$major Indian cities, collected from the Central Pollution Control Board (CPCB) for 2024. The dataset incorporates air pollutants, combined with daily and hourly temporal features. After preprocessing, rigorous cleaning and feature engineering, the models were trained for both regression ($\mathbf{P M}_{2.5}$concentration prediction) and classification (Air Quality Index category). Among regressors, LightGBM achieved superior performance with an$\mathbf{R}^{\mathbf{2}}$of 0.8649, RMSE of 21.97 and an MAE of 8.63. The voting ensemble classifier obtained the highest classification accuracy of 83.16 % with a macro F 1 -score of 0.77, performing well in the “Good” and “Satisfactory” categories but showing challenges in extreme classes due to imbalance. These findings suggest that the proposed system can enhance public awareness.