Spatial air quality index prediction model based on decomposition, adaptive boosting, and three-stage feature selection: A case study in China

Abstract

Accurate prediction of air quality index (AQI) plays a vital role in pollution control and air quality warnings. Considering the strong spatial correlation of air pollutants, this study focuses on the spatial prediction of AQI and proposes a novel hybrid model combining decomposition, three-stage feature selection, and boosted extreme learning machine. The spatiotemporal features are gradually extracted in three stages, including spatial correlation analysis, mutuality, and binary grey wolf optimization (BGWO). Moreover, decomposition and ensemble methods are adopted to further enhance the robustness and accuracy of the model. Based on maximal overlap discrete wavelet packet transform (MODWPT), the original AQI series in target site is decomposed into eight subseries with different frequency bands to reduce the non-stationarity of time series. Outlier robust extreme learning machine (ORELM) serves as base predictor, and the adaptive boosting. MRT (Adabost.MRT) is adopted to combine several ORELMs to generate more robust forecasts. To comprehensively evaluate the proposed model, data collected from 1497 air quality monitoring sites in China are used, and six sites are randomly selected as prediction targets. The proposed model is compared with several baseline models and existing models. The results demonstrated that the proposed model significantly outperforms other models in all target sites. All components of the proposed hybrid model are proved to be indispensable to guarantee outstanding performance. The spatiotemporal data are reasonably taken into consideration, which provides improvement for air pollution prediction models.