Class Imbalance SS-ELM for Regional Air Pollution Prediction

摘要

As one of the key environmental concerns, air pollution has aroused extensive attention recently. Many cities have built on-the-ground monitoring stations to measure hourly concentration of air pollutants, and there is an urgent demand for effective air pollution prediction model. However, most existing studies have mainly focused on improving prediction accuracy, which usually ignore the class imbalance problem and considerable cost for collecting labeled data. To promote the air pollution prediction performance, an innovative air pollution forecasting model is proposed based on class imbalance semi-supervised extreme learning machine (SS-ELM). First, majority weighted minority oversampling technique (MWMOTE), was employed to construct the balanced dataset. After preprocessing, SS-ELM is then selected to establish the model for classification and predict the air pollution level. Finally, extensive experiments were conducted on hourly pollutant data collected from Xiasha Economic Development Zone, Hangzhou, Zhejiang Province, China. The results demonstrate that compared with several existing methods, the proposed method has yielded superior performance in terms of G-mean and F-measure metric.