Prediction of daily PM2.5 concentration using support vector training combination (SVM) - adaptive and principal component analysis (PCA)

Air quality control is an inevitable issue at the forefront of national concerns. The aim of this study was to predict the daily concentration of PM2.5.

According to the objective, the type of research can be considered practical, and the statistical population of the research includes meteorological and pollution measuring stations within the 22 districts of Tehran. However, the statistical sample (synoptic geophysical station and Tarbiat Modares measuring station) was selected using a non-random sampling method. The desired statistical year for the study included the daily data from the selected stations for one year. Eleven input variables were used, which included meteorological data from the geophysical synoptic station (maximum and minimum temperature, minimum and maximum relative humidity, rainfall, maximum wind speed, and wind direction) and pollution data of PM2.5 concentration from the Tarbiat Modares station (daily concentrations of PM2.5 and the previous day). The support vector machine (SVM) model was used for prediction in this step.

The model was able to predict the daily concentration values of the PM2.5 pollutant for the upcoming days with a detection coefficient R² = 0.611 and RMSE = 10.87. In the second method, the support vector machine (SVM) model was combined with principal component analysis (PCA) to reduce the number of variables and perform modeling.

The results of this study show that the performance of the combined model is superior to the previous model, as the coefficient of determination R² increased to 0.65 and the error value decreased to 10.37 RMSE (root mean square error). This hybrid model (PCA-SVM) can assist city managers and decision-makers in controlling and reducing the amount of PM2.5 pollutants.