Table 2 Station-level and city-level OODG.

From: Exploration of transfer learning techniques for the prediction of PM10

 

City-level

Station-level

Stations

N-E-S-W-D

E-S-W-D

N-S-W

N

E

S

W

D

NRMSE

9.32

9.33

9.71

10.26

10.87

9.57

10.24

9.53

Avg. NRMSE

9.44

10.13

 

City-level with PM10_lag

Station-level with PM10_lag

Stations

N-E-S-W-D

E-S-W-D

N-S-W

N

E

S

W

D

NRMSE

7.15

7.13

7.47

8.31

8.01

7.57

8.14

7.39

Avg. NRMSE

7.25

7.88

  1. The outcomes of city-level RF models, trained on data from various stations in Graz, and station-level RF models, trained on data from a single station in Graz, are presented in the context of experiments conducted on the Zagreb station. Normalized root mean square error (NRMSE) serves as the performance metric. To assess the impact of PM10_lag features on the prediction performance, the second part of the table shows models including lagged values. In the city-level approach, performance remains relatively stable across directions (N-E-S-W-D and E-S-W-D), with minor distinctions observed, notably, N-S-W performing slightly inferior. Conversely, the station-level approach exhibits higher performance fluctuations; specifically, training on Graz station E and testing on Zagreb yields significantly inferior results compared to training on Graz station S and testing on Zagreb.
Back to article page