Page 79 - Çevre Şehir İklim İngilizce - Sayı 4

Hacı Abdullah Uçan - Tayfun Dede - Sinan Nacar

Table 16. Performance statistics of the training, verification and test data sets of
ANN-based models developed for the surface area variable
Training ANN_5 ANN_10 ANN_15 ANN_20 GWO_5 GWO_10 GWO_15 GWO_20

RMSE
(decare) 14806 14209 13832 12221 28593 43188 16110 17356
MAE 8755 8783 8290 7794 25664 39532 10895 11129
(decare)
NS 0,47 0,51 0,54 0,64 -0,98 -3,51 0,37 0,27

Verification ANN_5 ANN_10 ANN_15 ANN_20 GWO_5 GWO_10 GWO_15 GWO_20

RMSE 9908 10123 9578 10224 28095 42245 8423 9892
(decare)
MAE 6629 8533 7455 8247 25601 38626 5846 7214
(decare)
NS 0,50 0,48 0,54 0,47 -3,00 -8,04 0,64 0,50

Test ANN_5 ANN_10 ANN_15 ANN_20 GWO_5 GWO_10 GWO_15 GWO_20

RMSE 9954 10239 8548 10706 30266 43342 10804 11642
(decare)
MAE 7610 7608 6651 7803 27093 40588 8745 9773
(decare)
NS 0,54 0,51 0,66 0,47 -3,27 -7,75 0,46 0,37

When Table 16 is examined, it is seen that the performance statistics of
the GWO_15 model give good results in verification data sets, but from
the point of view of training and test data sets, lean ANN models exhibit
better performances. Of the models developed using different numbers of
neurons, it has been determined that the model with the highest accuracy
is the ANN_15 model, which was established using 15 neurons. As a general
evaluation, when regression-based models and ANN-based models are
compared, it is seen that the model developed using the TreeNet method
shows higher performance compared to the ANN models. In order to make
this comparison clearer, time series of the surface area variable have been
prepared (Figure 7).

68 The Journal of Environment, Urban and Climate