Multi-Resolution Wavelet Enhanced Hybrid Deep Learning Framework for Air Quality Index Prediction

Air quality is a crucial issue to monitor and predict in order to safeguard human health and guide urban planning. Nevertheless, due to the variability and complexity of environmental data, they pose significant challenges to the accuracy of prediction. In this paper, the authors propose a superior forecasting approach that combines the use of wavelet-based feature improvement and hybrid deep neural networks to forecast Air Quality Index (AQI). The data set includes the readings of the AQI and the pollutant levels on a daily basis between January 2019 and December 2024 at a monitoring station that is situated within Dwarka, Delhi. The input variables include the atmospheric conditions, such as Air Temperature (AT), Relative Humidity (RH), Wind Speed (WS), Wind Direction (WD), Solar Radiation (SR), Barometric Pressure (BP) and air pollutants, such as PM2.5, PM10, NO, NO₂, NO_X, SO₂, CO, Ozone, Benzene, Toluene. Eight hybrid deep learning designs are used to test the effectiveness of standalone and wavelet-integrated design: CNN-BiGRU, CNN-BiLSTM, CNN-BiLSTM-BiGRU, W-CNN-BiGRU, W-CNN-BiLSTM, W-CNN-BiLSTM-BiGRU, W-CNN-BiGRU-BiLSTM. Fine-tuning and comparing of models are done by using performance indicators such as Correlation Coefficient (R²), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). The findings indicate that wavelet and enhanced networks note significant improvements in networks that are not decomposed, and W-CNN-BiGRU-BiLSTM network is the most predictive with a high =0.9853 and low error metrics of RMSE= 25.58, MAE= 18.64 and MAPE= 6.43%. The findings confirm the fact that the combination of multi resolution analysis and hybrid neural networks is a prospective future in enhancing AQI prediction and enabling proactive environmental decision making.