Integrating machine learning techniques for Air Quality Index forecasting and insights from pollutant-meteorological dynamics in sustainable urban environments

Air pollution poses a significant environmental and health challenge in Delhi, India. This research focuses on predicting the Air Quality Index (AQI) for Delhi utilizing machine learning techniques. The research methodology encompasses comprehensive steps such as data collection, preprocessing, analysis, and modeling. Data comprising various pollutants and meteorological parameters were gathered from the Central Pollution Control Board (CPCB) spanning from January 1, 2016, to December 30, 2022. Missing values were imputed using the IterativeImputer method with RandomForestRegressor as the estimator. Data normalization and variance reduction were achieved through Box-Cox transformation. Spearman Rank Correlation analysis was employed to explore relationships between features and AQI. Initial evaluation of nine machine learning algorithms identified Random Forest and XGBoost as the top performers based on accuracy. These algorithms were further optimized using 5-fold cross-validation with RandomizedSearchCV. The results demonstrated the efficacy of both algorithms in AQI prediction. Notably, PM 2.5 and CO concentrations emerged are most influential features, highlighting the potential for AQI improvement in Delhi through the reduction of these pollutants. This research distinguishes itself through a meticulous examination of the complex interconnections between pollutants and AQI, providing invaluable insights to inform targeted interventions and enduring policies geared towards improving air quality in Delhi.