Page 61 - Çevre Şehir İklim İngilizce - Sayı 4
P. 61

Hacı Abdullah Uçan - Tayfun Dede - Sinan Nacar


            and population were obtained from the relevant institutions and organizations
            in quarterly periods from 2002 to 2020 in order to create forecast models for
            the urban transformation process in Türkiye. These data used is available in
            the study of Uçan, 2022. The prediction models of the urban transformation
            process were examined with regression-based and ANN-based modeling. In
            regression analyses, KRA: Classical Regression Analysis, MARS: Multivariate
            Adaptive  Regression  Splines  (Friedman,  1991),  TreeNet  Gradient  Boosting
            Machine (Friedman, 2001) were used, and in the ANN model, GWO: Gray
            wolf optimization algorithm (Mirjalili et al. 2014) was coded in the MATLAB
            programming language and added to the ANN.
               While starting the studies on estimating the dependent variables of the
            number  of  flats,  number  of  licenses  and  surface  area,  the  independent
            variables  that  may  have  major  effect  on  these  variables  were  determined.
            At this stage all possible regression analyses were performed to determine
            which one of these variables are more effective and it was identified that the
            model using all variables has the highest performance. In this context, gross
            national  product  (GNP),  inflation  rate  (%),  interest  rate  (%)  and  population
            were determined as independent variables. In addition to these variables, the
            number of periods in each year was included in the independent variables as
            another  independent  variable.  These  independent  variables  were  used  for
            all three dependent variables without modification. Modeling studies were
            started by using three different dependent variables against five independent
            variables in total. In order to test the accuracy of the models established in
            the modeling studies, the data sets are divided into training, verification and
            test data sets. In this way, the accuracy of the models established with the
            training  and  verification  data  sets  could  be  evaluated  using  the  test  data.
            The performances of the models, the root mean squared error (RMSE), mean
            absolute  error  (MAE),  and  Nash-Sutcliffe  (NS)  performance  statistics  were
            calculated using the following equations.
               RMSE=                      (1)



               MAE=                       (2)


               NS=                        (3)


               In the given equations; t  indicates the measurement values, t di   shows the
                                    i
            forecast values, the average of the measurement values and the number of n
            data. The performance of the smallest value among RMSE and MAE values is
            evaluated as high, while the NS value is between −∞ and 1. NS=1 indicates
            that the method used is excellent.



             50  The Journal of Environment, Urban and Climate
   56   57   58   59   60   61   62   63   64   65   66