Timeseries horizon prediction question

Liliput
Liliput New Altair Community Member
edited November 5 in Community Q&A
Hi all,

I am working on a POC project and have the following setup:
  1. dataset has the structure [DateTime, var1, var2, var3, target] - all real values except DateTime of course with 1min sample rate
  2. I am creating a window with params: window size = 60(1hour), step size = 15, horizon size = 1, horizon offset = 14 so I want to be able to predict the value of "target" in 15min.
  3. I use Auto Model with the windowed data
All is working fine and I get ~1.5% relative error which is good I assume. The problem I have and do not understand is: 

 - Why if I plot the "traget +15 (horizon)" as a line in the time it follows "target" almost exactly and in particular is not 15min shifted in time as I am expecting it to be? I can understand why if I am predicting the value in X mins my predictions should be 15min ahead so if I plot the predicted attribute it should approximate target but have a phase of 15min in time.

Any help is appreciated to resolve my confusion.
Thanks   
Tagged:

Answers

  • lionelderkrikor
    lionelderkrikor New Altair Community Member
    Hi @Liliput,

    In order we can reproduce and understand what is going on, could you please share : 
     - your process
     - your data

    Regards,

    Lionel

Categories