Page 640 - NGTU_paper_withoutVideo

Modern Geomatics Technologies and Applications

Depending on the concentration of pollutant PM2.5 and using the below table, the concentration of pollutant is converted
to pollution class[14].

Table 2 PM2.5 threshold
AQI AQI PM2.5 pollutant
classification range breakpoints

Clean
0-50 0-12

Healthy 51-100 12.1-35.4

Unhealthy for
sensitive groups 101-1550 35.5-55.4

Unhealthy 151-200 55.5-150.4

Very unhealthy 201-300 150.5-350.4

Dangerous 301-400 350.5-9999.9

Accordingly, having the pollutant class and the available parameters, a structural table should be generated as follows:

Table 3 Structure of input and output data

Input output

moy 1 t dow 2 t hod 3 t xyz wd 4 t ws 5 t p 6 t 7 t t u 8 t 9 t r dI 10 t - 2 NN t class t

After creating a structure like Table 3, it is observed that some fields of the table are empty and in some cases the data
fluctuations are very high. Therefore, the Fourier series method[15] was used to fill the gap and the Savitzky-Golay[16] method

was used to soften the data. Next, the C4.5 and CART decision tree algorithms and comparing the accuracy of the methods have
been accomplished. Also, in this study, the overall accuracy and Kappa index have been used to express the accuracy. Overall
accuracy and Kappa index are statistical criteria to express the reliability between variables to recognize the quality of data,
which are calculated according as follows[17, 18]:

+
= (2)
+ + +

1 Month of year
2 Day of week
3 Hour of day
4 Wind direction
5 Wind speed
6 Air pressure
7 Temperature
8 Humidity
9 Rain
10 Inversion intensity
4