I have a multivariate time series with three features x1,x2,x3, I have chosen window size 5 to make a prediction on variable x3 at time t=6. Is it correct to use only the LSTM input on the variables x1,x2 to predict the value of x3 at time t=6???
1
-
1$\begingroup$ We don't have enough information to provide an answer. What is the problem at hand ? did you find some source or existing literature ? $\endgroup$– Lucas MorinCommented Jun 5, 2023 at 10:27
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
No.
You have 3 features, $x_1$, $x_2$, and $x_3$. For simplicity, I'm going to rename these features $a_t$, $b_t$, and $c_t$ respectively. Since this is a prediction model, the inputs to your neural network would be $a_{t-1}$, $b_{t-1}$, and $c_{t-1}$ and your output would be $c_t$. Therefore, you would not just use $a_{t-1}$ and $b_{t-1}$ to predict $c_t$, you would also use $c_{t-1}$. A feature is an input to a neural network, and you cannot disregard an input.
