Data and Knowledge: An Interdisciplinary Approach for Air Quality Forecast

Cheng Feng, Wendong Wang, Ye Tian, Xiangyang Gong, Xirong Que

Abstract: Air pollution has become a critical problem in rapidly developing countries. Prior domain knowledge combined with data mining offers new ideas for air quality prediction. In this paper, we propose an interdisciplinary approach for air quality forecast based on data mining and air mass trajectory analysis. The prediction model is composed of a temporal predictor based on local factors, a spatial predictor based on geographical factors, an air mass predictor tracking air pollutants transport corridors and an aggregator for final prediction. Experimental results based on real world data show that the cross-domain data mining method can significantly improve the prediction accuracy compared with other baselines, especially in the period of severe pollution.