Page 82 - Çevre Şehir İklim İngilizce - Sayı 4

Urban Transformation Forecasts with Artificial Intelligence
Based Algorithms
Conclusions

Statistical values between the years 2002-2020 were obtained from the
relevant institutions for the prediction models. In order to allow sufficient
performance of the prediction models, a mixed selection was made in the data
sets in the order described in the studies. Linear, antilogarithm, exponential
and quadratic functions as well as TreeNet and MARS methods were used
in traditional regression-based models. In ANN-based models, the gray wolf
optimization (GWO) algorithm, a meta-intuitive algorithm based on swarm
intelligence, was used to optimize coefficients and the lean ANN in different
neuron numbers. As a result of measuring the performance values of the
prediction models, it was determined that good estimates were made.
In the modeling carried out to estimate the number of flats, the TreeNet
method gave the closest estimates to the actual values. According to the
performance values of the prediction models, the TreeNet model has given
good results and the ANN model has given very good results. Much more good
estimations were made with an improvement of 27% compared to the best
estimation obtained from the ANN-based model regression-based modelling.
In the modeling carried out to estimate the number of licenses, the TreeNet
method gave the closest estimates to the actual values. According to the table
measuring the performance values of the prediction models, good results have
also been obtained for the MARS method. Much more good estimations were
made with an improvement of 10% compared to the best estimation obtained
from the modelling performed with MARS and TreeNet method.
In the modeling carried out to estimate the surface area, the TreeNet
method gave the closest estimates to the actual values. According to the
table measuring the performance values of the prediction models, the
TreeNet model has given good results and the ANN model has given very
good results. Much more good estimations were made with an improvement
of 24% compared to the best estimation obtained from the artificial neural
network-based model regression-based modelling.
When the estimation models for the number of apartments, the number of
licenses and the surface area are generally considered, the ANN method has
given better estimates than the traditional regression methods. When the
prediction models are examined in detail for the training, validation and test data
sets, it is seen that some models obtained by adding gray wolf optimization to
the ANN yielded better results than the models established with other methods.
In the analyses of the model obtained by adding the gray wolf optimization
to the ANN, the number of populations was taken into account as 100, the
maximum number of iterations as 1000, and the lower-upper limit values as
-5 and +5. In order for this type of model to give better prediction results, a

Year 2 / Issue 4 / July 2023 71