1. Random forest regression for statistical modeling and forecasting of PM10.
- Author
-
Ivanov, A., Gocheva-Ilieva, S., and Stoimenova-Minova, M.
- Subjects
- *
STATISTICAL models , *REGRESSION analysis , *RANDOM forest algorithms , *BOX-Jenkins forecasting , *PETROLEUM - Abstract
The PM10 (Particulate Matter with diameter of 10 microns or less) is a major air pollutant with a number of harmful effects on human health. As a primary pollutant, it is also an indicator of the overall level of the ecological state of the environment. This determines the need and appropriateness of research on the accumulated empirical data, especially in affected areas. This paper examines the average daily PM10 levels in a specific region of Bulgaria, near the Black Sea coast and a large refinery for oil and petroleum products. The data of PM10 and meteorological variables such as air temperature, humidity, wind speed and others for a period of more than 6 years are studied. Using the Box-Jenkins ARIMA model, an autoregressive term was identified in the PM10 time series, which was used as an additional predictor in the models. Regression analysis with the Random Forest (RF) machine learning method is used for statistical modeling of the time series. RF models were obtained describing the data by almost 94%. The models are applied for short-term forecasts of PM10 pollution with 1 to 7 days ahead. The comparison with the actual measurements showed that the proposed approach gives very good results and could be embedded in mobile software for air pollution forecasting. [ABSTRACT FROM AUTHOR]
- Published
- 2022
- Full Text
- View/download PDF