Search for a command to run...
Air pollution is found to be the foremost constraint for urbanization in developing countries. The vital indicator to assess the levels of air pollution is termed to be the Air Quality Index (AQI). Accurate AQI predictions are crucial for effectual environmental supervision and introduce any new public health initiative. This research introduces an enhanced model for AQI prediction by means of a hybrid approach which integrates time-series analysis and deep learning. The primary data for this study was collected from the Central Pollution Control Board (CPCB) India, including an extensive range of pollutants measured across various cities over an extended period. We use the Transductive Long Short-Term Memory (TLSTM) network with a Seasonal Autoregressive Integrated Moving Average (SARIMA) model which captures seasonal patterns and long-term dependencies prevalent in the data. The proposed Hybrid Model’s TLSTM incorporates a time decay mechanism to account for the temporal dynamics of air pollution, which enhances the model ability in handling unstable time intervals between data points. Additionally a Convolutional Neural Network (CNN) layer extracts local patterns from the data, and a bidirectional RNN is employed to capture context from both previous and upcoming time steps. The final predictions are refined using an XGBoost boosting stage, which combines linear components of pollutants with the TLSTM predictions. The Hybrid Model demonstrates an exceptional predictive performance when compared to the existing SARIMA-LSTM model. The Root Mean Squared Error (RMSE) value is reduced from 0.179 to 0.02196 (12.2%) and the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value is increased from 0.512 to 0.9896(47.76%). These results emphasize the high accuracy and reliability of the proposed model in AQI value prediction. The proposed approach not only outperforms traditional models but also provides potential insights into the seasonal and temporal variations of air pollutants.