The data set was created by combining the collected data in a single
center. Long short-term memory (LSTM) algorithm is preferred in the
model developed to predict the classification type according to
temperature and humidity values. LSTM is a recurrent neural network
(RNN) architecture that remembers values at random intervals [32].
An LSTM is well suited for classifying, processing and predicting time
series given the time delays of unknown size and duration between
significant events [33].
Basically, the internal structure of the LSTM architecture; It consists
of input (Equation 4), forget (Equation 5) and output gates (Equation 7)
and input layer (Equation 6). In the LSTM architecture, first of all, π₯π‘
and βπ‘β1, which are used as inputs, are decided which information will
be deleted. This is done by a sigmoid layer (Equation 3) called the
forget the door layer [34].
\(\sigma\left(x\right)=\left(1+e^{-x}\right)^{-1}\) (3)
\(i_{t}=\sigma\left(W_{i}x_{t}+R_{i}h_{t-1}+b_{i}\right)\) (4)
\(f_{t}=\sigma\left(W_{f}x_{t}+R_{f}h_{t-1}+b_{f}\right)\) (5)
\(g_{t}=tanh\left(W_{g}x_{t}+R_{g}h_{t-1}+b_{g}\right)\) (6)
\(o_{t}=\sigma\left(W_{o}x_{t}+R_{o}h_{t-1}+b_{o}\right)\) (7)
In the LSTM model, the first step after dataset partitioning is to
decide which information to discard in the cell. This decision is made
by the sigmoid layer called the forget gate layer in the LSTM model. It
is then decided which new information will be stored in the cell state.
For this, the input gate layer is used. A vector of the new values ββto
be generated is also produced with the tanh layer. The output layer is
created by combining the generated information and vectors. This output
is filtered by cell state. Here, first, a SoftMax layer is run, which
decides which parts of the cell state it will output. The SoftMax layer
extracts the ordered order classifications.
In the experiments conducted to develop the proposed LSTM model and
evaluate its performance; Intel (R) I9 3.2 Ghz processor hardware with
64 GB RAM and Python programming language is used in the Spyder
interface. The dataset was randomly partitioned as 80% for training and
20% for testing. The hyperparameters used in the LSTM model are given
in Table 3.
Table 3. Hyperparameters and values of the model