Comparison of Machine Learning Methods for Air Quality Prediction Based on Particulate Matter (PM2.5)

Particulate PM2.5 has a major effect on human health compared to other types of pollutants such as NO2, 03, CO, and SO2. PM2.5 is the worst type of major pollutant for human health that can cause various types of serious diseases and even death. PM2.5 is defined as an ambient air particle, which is one of the particle elements found in the free air of the earth's surface in the troposphere layer, which is very small in size, reaching <2.5 microns which contains causing substances include polycyclic aromatic hydrocarbons, lipopolysaccharides, and transition metals (copper, manganese, and zinc). Prediction of air quality index values in national cities can help the government make policies to carry out air pollution prevention and mitigation measures by looking at local air quality trends in advance, especially when there is a significant increase in the amount of PM2.5 pollutants. The method of predicting air quality index values can be done using machine learning techniques with time series data analysis. This study will compare machine learning methods with two approaches, namely statistical regression with RF and ARIMA and deep learning method, namely LSTM. Based on the research results, the best machine learning method approach is statistical regression, namely the RF method with accuracy evaluation using RMSE, MAE, and MSE, which are respectively with a minimum value of 0.162, 0.108, and 0.026, followed by the LSTM and ARIMA methods.