Page 66 - Çevre Şehir İklim İngilizce - Sayı 4

Urban Transformation Forecasts with Artificial Intelligence
Based Algorithms
Results Related to the Number of Flats

The performance statistics obtained for the training, verification and test
data sets from regression-based classical regression analysis (KRA), MARS and
TreeNet methods established to estimate the number of flats variable using
mixed selection data sets are given in Table 2.
Table 2. Performance statistics of the training, verification and test data sets of
regression-based models developed for the number of flats variable

Training KRA_Lin KRA_Üs KRA_Eks KRA_Kua MARS TreeNet
RMSE(pcs) 74728 74235 74565 57020 67946 65507

MAE (pcs) 47777 46577 46969 39392 42630 25786
NS 0.44 0.45 0.44 0.67 0.54 0.57
Verification KRA_Lin KRA_Üs KRA_Eks KRA_Kua MARS TreeNet

RMSE(pcs) 46254 53487 47512 69126 37824 39377
MAE (pcs) 36579 41567 38250 55432 31454 33492

NS 0.49 0.32 0.46 -0.14 0.66 0.63
Test KRA_Lin KRA_Üs KRA_Eks KRA_Kua MARS TreeNet
RMSE(pcs) 47737 44430 46062 96007 48943 40256
MAE (pcs) 38703 34161 35286 80720 43746 34778
NS 0.52 0.59 0.56 -0.93 0.50 0.66
Lin: Linear, Üs: Antilogarithm, Ex: Exponential, Kua: Quadratic
When Table 2 is examined, it is seen that the lowest RMSE and the highest
NS values for the training data set are obtained from the KRA_Kua model, and
the lowest MAE value is obtained from the TreeNet model. However, looking
at the validation and test data sets, it is understood that the performance of
the KRA_Kua model is poor compared to other models. In addition, taking
into account the verification and test data sets it is seen that the highest
performance for the verification data set is obtained from the MARS method,
while the highest performance for the test data set is usually obtained from
the TreeNet method. As a general evaluation, it can be said that the TreeNet
method gives high accuracy results for all three data sets. In the KRA method,
it was determined that the model performances of linear, antilogarithm and
exponential functions are close to each other, and the exponential function is
the one with the highest accuracy among them. The coefficients of the models
developed with linear, antilogarithm, exponential and quadratic functions are
given in Table 3.

Year 2 / Issue 4 / July 2023 55