Improving PM 2.5 Prediction Accuracy with a Hybrid EEMD-CNN-BiLSTM Approach

Due to its detrimental impact on both the environment and human health, air pollution is a major global health concern. One of the main causes of Delhi’s air pollution is particulate matter (PM 2.5), and exposure to it, whether short-term or long-term, can have detrimental effects on one’s health. Effective prediction of PM 2.5 is crucial for improving accuracy and reducing the impact of severe pollution events, which enhances decision-making related to environmental management. However, accurate forecasting of PM 2.5 levels is challenging due to the unpredictable nature of the data. In this proposed strategy, EEMD based decomposition algorithm is employed to de-noise the PM 2.5 data. EEMD decomposes the pollution data into 9 IMFs and one residue component. Then, the CNN-BiLSTM predictive model is apply to each extracted IMF as well as the residue component. Afterwards, the predicted results of each component obtained using CNN-BiLSTM are added to determine the final forecast. To validate the effectiveness of the hybrid method, its performance is compared against other models, such as traditional BiLSTM and conventional CNN-BiLSTM, using statistical error metrics including MAE, RMSE and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. The proposed model yields an MAE of 7.77, an RMSE of 11.86, and an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.99. The empirical results demonstrate that the proposed hybrid approach, based on EEMD-CNN-BiDLSTM, outperforms traditional methods in terms of prediction accuracy.