A Comparative Analysis into the performance of Artificial Intelligence models for predicting the Air Quality Index- Case study India

Air pollution is a perilous growing global issue with profound public health and ecosystem impacts globally. Delhi, India, is one of the regions at the forefront of this issue, facing some of the highest AQI levels in India, with it being ranked the most polluted capital in the world by IQAir, an international monitoring agency. This paper looked at predicting daily average AQI index values in Delhi using Artificial Intelligence Machine Learning models and compared the performances of each models: Random Forest Regressor, N-BEATS, ACNN, CatBoost, XGBoost, etc. which were trained on a year-round as well as season-specific AQI data of Delhi, India. The data, sourced from the CPCB, contained 5 complete years of AQI data. For the year-round datasets, the N-BEATS model, with an R2 and RMSE of 0.83 and 47.70 respectively; GRU with MSE loss function, with an R2 and RMSE of 0.85 and 43.59 respectively; and Random Forest Regressor, with an R2 and RMSE of 0.86 and 45.54 respectively, had the best performance; however, it was observed that the Random Forest Regressor, Catboost and XGBoost models performed consistently well for all datasets. The comparison presented Random Forest Regressor model with the highest performance across datasets with the highest average R2 and lowest average RMSE of 0.7360 and 55.2720 closely followed by CatBoost and XGBoost. Overall, the results show that the Random Forest Regressor model performs the best when it comes to predicting the Air Quality Index in Delhi, India.