Page 78 - Çevre Şehir İklim İngilizce - Sayı 4
P. 78

Urban Transformation Forecasts with Artificial Intelligence
                                        Based Algorithms

               Basic  functions  obtained  from  MARS  method  and  equation  used  for
            achieving the surface area are given in Table 14. The equation used to achieve
            the variable surface area is available in the last row of the table. The relative
            importance of the independent variables used in the equations achieved from
            the MARS and TreeNet methods are also given in Table 15.


                Table 14. The equation used for achieving the surface area variable and basic
                             functions obtained from the MARS method
            Basic functions
            BF1 = max( 0, N - 0,753556); BF2 = max( 0, 0,753556 - N);
            BF5 = max( 0, G - 0,1);
            BF6 = max( 0, G - 0,682458) * BF2; BF10 = max( 0, 0,148466 - F)
            * BF2; BF12 = max( 0, 0,128509 - E) * BF2;
            F_yüz_ölçümü = 0,307045 - 1,53276 * BF1 - 0,291522 * BF2 + 0,289436 * BF5 - 2,55668 * BF6 +
            26,492 * BF10 - 1254,41 * BF14;

                Table 15. The relative importance of the independent variables used for the
                   estimation of the surface area variable obtained from the MARS and
                                       TreeNet methods
               MARS                               TreeNet
              E   100   ||||||||||||||||||||||||||||||||||||||||||||||||  N 100  ||||||||||||||||||||||||||||||||||||||||||
                                                 G 96,67    |||||||||||||||||||||||||||||||||||||||||
                                                 E   83.16  |||||||||||||||||||||||||||||||||||
                                                 F   70.18  |||||||||||||||||||||||||||||
                                                 Y   57.64  ||||||||||||||||||||||||
               Table 15 shows that population, only the inflation rate was effective among
            the independent variables in the equations obtained for the MARS method,
            while all variables were effective in the equation achieved for the TreeNet
            method.  The  variable  with  the  highest  relative  importance  for  the  MARS
            method  was  the  inflation  rate  variable,  while  the  variable  with  the  highest
            relative  importance  in  the  TreeNet  method  was  the  population  variable.
            The performance statistics obtained for the training, verification and testing
            datasets of ANN-based models established to predict the surface area variable,
            including peak values, are given in Table 16 for four different neuron numbers.
            Tests were conducted by trial and error for the neuron numbers given in the
            table, and the results of the models that gave the highest performance values
            were presented within the scope of the thesis.






                                                                 Year 2 / Issue 4 / July 2023  67
   73   74   75   76   77   78   79   80   81   82   83