Multi‐step air quality index forecasting via data preprocessing, sequence reconstruction, and improved multi‐objective optimization algorithm

This research presents a hybrid model for multi‐step, interval forecasting of air quality indices. An efficient preprocessing module is applied to split the raw data into various sub‐series, and the optimal mode of data input is determined through feature selection. A multi‐objective optimization algorithm is proposed to tune the parameters of kernel extreme learning machine to achieve high accuracy and stability. An evaluation with several error criteria, benchmark models, and critique is conducted using three daily air quality index datasets from three cities of China. Empirical results indicate that the developed model achieves superior predictions of air quality indices, which may be useful in policies for mitigating air pollution.