A comprehensive evaluation of air pollution prediction improvement by a machine learning method

Urban air pollution prediction is one of the most important tasks in the treatment of urban air pollution. Due to the disadvantage that source list updated not in time for WRF-Chem which is a numeric model, the prediction result may be not good enough. In this paper, we take full advantages of forecast on pollution, weather, chemical component from WRF-Chem model as input features, design a comprehensive evaluation framework to improve the prediction performance. Experiments are implemented with different features groups and classification algorithms in machine learning method for 74 cities in China, to find the best model for each city. From experiments, for different city, the best result can be obtained by different group of feature selection and model selection. Experimental results indicate that the more feature we used, the more possibility to enhance the accuracy. For method aspect, the result from combined model is better than the unique model.