Ensemble multifeatured deep learning models for air quality forecasting

Abstract

As air pollution becomes increasingly serious, accurate forecasting of air quality has become an important issue. Many studies related to machine learning and deep learning methods have been proposed to predict air quality. These studies have also confirmed that deep learning has better predictive performance than traditional machine learning. Ensemble Learning has also been proven that can improve the classification efficacy of machine learning. In order to improve the accuracy of air quality forecasting, in this paper, we first exploit a GRU (Gated Recurrent Unit) deep learning network architecture to design various predictive models (GRU13d, GRUAW13d, GRUAW14d, GRUSS13d, and GRUST14d, etc.) that meet various spatial and temporal situations, and then propose an ensemble learning forecasting model, named MLEGRU (Multiple Linear Regression based GRU), based on multiple linear regression technique to integrate these deep learning predictive models. We utilized 67 monitoring stations of EPA (Environmental Protection Administration) in Taiwan to train and test the forecasting models with data collected from years 2013–2019, and use actual forecasted results to evaluate the performance of MLEGRU. The results are compared with other ensemble methods by using various factors, such as MAE (Mean Absolute Error), RMSE (Root Mean Square Error), and AEL3 (Absolute Error less than 3). The results reveal that the proposed MLEGRU model provides a better forecasting accuracy than other ensemble learning methods.