How do you represent multivariate multistep data using traditional machine learning? I know this seems like a tailored problem for RNN/LSTM, but I am wondering what the alternative machine learning solution would be.


time, feature_1, feature_2, feature_3
1     25         150        0.7       
2     25.3       147        1.1        
3     25.5       145        0.8
10    26.7       165        1.2

Now I want to predict feature_1 at time 10 using the first 3 time steps of all features, and get a dataset like this

feature_1,       feature_2,       feature_3,       target
[25,25.3,25.5]   [150,147,145]    [0.7,1.1,0.8]    26.7
  • $\begingroup$ Would you mean something like vector autoregression? $\endgroup$
    – Dave
    Jun 10 '20 at 10:08

You can always reduce the forecasting problem to a tabular regression problem and then apply any tabular regression algorithm you like. Here's a good explanation of how this works.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.