I am trying to do a time series forecasting with multiple variables.
As you can see in the figure attached, I have 3 variables, including the variable to predict which is Feature 3. I am trying to predict Feature 3 one time step ahead using all the variables from the previous two time steps. So I am clear that the input shape will look something like this [timesteps, number of features] = [2,3].
Now, I am having some difficulty in how to reshape my dataset in the same format like my table below. I appreciate if someone can help me with this.
what you are looking for is the Lag Column node.
It allows you to create shifted copies of your columns.
In your title you mention LSTMs but for what you are describing an LSTM actually won’t work very well since its strength is that it can learn long time dependencies (at least better than many other ML approaches).
If you know exactly how many time steps you want to look back, I would recommend other algorithms like a simple Logistic Regression, Decision Trees, Gradient Boosted Trees or Random Forests.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.