60
H. Altıparmak et al.
9.4.1 Training Sets and Training Process
The training process is not random and must be planned in advance. The input data
should not be randomly transmitted to the artificial neural network. Preparing the
data to be trained in the network in advance and entering the network in sequential
order provides better results. The regular data set for training the network is called the
training set, which should contain as much data that the network may encounter later
as possible. After the network has been trained, acceptable answers can be received
from the network when a series of data is entered in the network that is not included
in the training set. This capability of the network increases in direct proportion to
the diversity of data in the training set.
9.4.2 Scaling of Input and Output Data
In a neural network that uses a sigmoid function, the network can be given numbers
between 0 and 1 or only between these two values. However, the network may be
required to use large number to achieve this. The input and output data between any 2
values are multiplied by two coefficients. Thus, as the network continues to operate
with values between 0 and 1, the input-outputs can be exchanged at the desired
intervals.
9.4.3 Minimum–Maximum Normalization (Min–Max
Normalization)
In this method, the largest and smallest values in a group of data are handled. All
other data is normalized to these values. The aim is to normalize the smallest value
to 0 and the largest value to 1, and to spread all other data over this 0–1 range.
Various studies have documented the increase in the average global temperature
in the last century. The consequences of a continuous rise in global temperature will
be significant. Rising sea levels and increased frequency of extreme weather events
will affect billions of people.
The dataset contains data from the past 25 years.
9.5 Neural Network Performance
We used a data table comprising 8 rows and 303 columns as input. We used a feedback
neural network consisting of 1 hidden layer and 10 neurons (Fig. 9.1).
H. Altıparmak et al.
9.4.1 Training Sets and Training Process
The training process is not random and must be planned in advance. The input data
should not be randomly transmitted to the artificial neural network. Preparing the
data to be trained in the network in advance and entering the network in sequential
order provides better results. The regular data set for training the network is called the
training set, which should contain as much data that the network may encounter later
as possible. After the network has been trained, acceptable answers can be received
from the network when a series of data is entered in the network that is not included
in the training set. This capability of the network increases in direct proportion to
the diversity of data in the training set.
9.4.2 Scaling of Input and Output Data
In a neural network that uses a sigmoid function, the network can be given numbers
between 0 and 1 or only between these two values. However, the network may be
required to use large number to achieve this. The input and output data between any 2
values are multiplied by two coefficients. Thus, as the network continues to operate
with values between 0 and 1, the input-outputs can be exchanged at the desired
intervals.
9.4.3 Minimum–Maximum Normalization (Min–Max
Normalization)
In this method, the largest and smallest values in a group of data are handled. All
other data is normalized to these values. The aim is to normalize the smallest value
to 0 and the largest value to 1, and to spread all other data over this 0–1 range.
Various studies have documented the increase in the average global temperature
in the last century. The consequences of a continuous rise in global temperature will
be significant. Rising sea levels and increased frequency of extreme weather events
will affect billions of people.
The dataset contains data from the past 25 years.
9.5 Neural Network Performance
We used a data table comprising 8 rows and 303 columns as input. We used a feedback
neural network consisting of 1 hidden layer and 10 neurons (Fig. 9.1).