A study on PM10 forecasting model using machine learning†
Sahm Kim1
1Department of Statistics, Chungang University
Correspondence to: † This research was supported by the Chung-Ang University research grant in 2022.
1 Professor, Department of Statistics, Chungang University, Seoul 06974, Korea. E-mail: sahm@cau.ac.kr
1 Professor, Department of Statistics, Chungang University, Seoul 06974, Korea. E-mail: sahm@cau.ac.kr
Received July 24, 2023; Revised August 7, 2023; Accepted August 11, 2023.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
- Abstract
- Fine dust refers to dust with a particle diameter of less than 10μg among dust, which is a particulate matter floating or flying down in the atmosphere, and is also referred to as PM10. These fine dust is very small in size and permeates the body without being filtered from the nose or bronchial tubes, causing inflammation through asthma, lung disease, or the action of immune cells. Recently, it was found that Korea has the highest concentration of fine dust in the world, and it is important to take measures through an accurate forecast system because fine dust directly affects not only health but also ecosystems and crops. Therefore, this paper attempted to compare machine learning prediction performance of fine dust concentration using weather data provided by the Korea Meteorological Administration and air pollutant data provided by Air Korea. As for the region, data from Incheon Metropolitan City, which is the closest to the Shandong Peninsula, the inflow path of yellow dust, were extracted, and a model was built after confirming the correlation between various weather factors and air pollutants in Incheon. MLP, RNN, LSTM, GRU, and CNN were used as models, and predictive performance was compared by organizing basic hyperparameters and single layers. After that, the GRU2 model, which added layers to the GRU1 (single layer) model, was newly constructed and compared with the GRU1 model with the best prediction performance. Prediction performance was evaluated by MAE and RMSE in test data. Most of them showed similar predictive performance, but it was confirmed that the GRU1 model had the best performance compared to other models, with MAE 8.80 and RMSE 14.61. The model with the lowest prediction performance was the MLP model, followed by RNN, LSTM, GRU2, and CNN.
- Keywords : CNN, GRU, LSTM, machine learning, PM10 forecasitng
Full Text(PDF) Free
