Stacked deep learning ensemble for time series prediction of PM2.5 levels in Bihar

A major contributor to the deterioration of air quality is Particulate matter (PM_2.5) with a diameter of less than 2.5 micrometers, which makes air pollution among the most pressing environmental concerns in the world. Bihar, one of India’s most densely populated states, has experienced deteriorating air quality over the past decade, particularly in major urban centers like Patna, Gaya, and Muzaffarpur. Given the severe health and environmental implications of high PM_2.5 levels, accurate forecasting models are essential for proactive pollution control measures. This study explores the application of various time series forecasting models for predicting PM_2.5 concentrations, focusing on deep learning ensemble methods. The framework utilizes five base deep learning models: Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Gated Recurrent Unit (GRU), and Bidirectional LSTM (Bi-LSTM), each trained individually to capture different aspects of temporal dependencies in the data. To enhance predictive accuracy, the predictions from these base models are combined using a stacking-based ensemble approach with XGBoost as the meta-learner. This ensemble method refines the final prediction by leveraging the strengths of each model and mitigating their individual weaknesses. The proposed model exhibits outstanding effectiveness in PM_2.5 estimation, attaining an MSE of 33.72, MAE of 2.56, RMSE of 5.80, and an R${}^2$ of 0.99 for Patna. Similarly, it attains an MSE of 8.90, MAE of 2.12, RMSE of 2.98, and an R${}^2$ of 0.99 for Gaya, while for Muzaffarpur, the model records an MSE of 11.37, MAE of 2.44, RMSE of 3.37, and an R${}^2$ of 0.99. These outcomes demonstrate how accurate and dependable the model is at predicting air quality in various places. The findings aim to assist policymakers and environmental agencies in making informed decisions to reduce air pollution and safeguard the general public’s health.