Predicting next integer value in a time series

jeton
jeton New Altair Community Member
edited November 2024 in Community Q&A
Hi!
I'm from Argentina and new to RapidMiner so I apologize in advance for my poor English.

I have a time series in an Excel that looks like this...

Date              Number
May-02        4
May-03        9
May-04        1
May-05        7
May-06        7

and so on... Numbers (they are integers from 0 to 9) can repeat and they are "random". The last record is, for example, Sep-18 is 2. Now... I know that Sep-19 is 7 but I can't find the right model and learner to achieve that.

Here are the steps:

- Read Excel
- Set Role (I labelled the date)
- Windowing (encode series by example, horizon 1, window size 1, step size 1, create single attributes, create label (for Number))
- Sliding Window Validation: Training (learner is SVM Evo, radial) and Testing (Apply Model, Forecasting Performance horizon 1)

Now when I make a copy of the first 3 of the nodes and create an Apply Model node in order to "compare" (with Windowing in horizon 0) the original date with the prediction label, the last record is 0!!!! And I kwow is 3! The rest of the "predictions" are OK.

Is there a simple way to make a prediction of the next value of an integer-valued time series based on the past values?
Hope I've been clear enough.
I really need your help guys!

Thank you very much!

Leo

P.S.: I tried to "copy" the way of modeling from this video... http://www.youtube.com/watch?v=UmGIGEJMmN8
Tagged:

Answers

  • jeton
    jeton New Altair Community Member
    Hi, goog morning!
    I don't know what to do, I'm stuck.

    The spreadsheet looks like this

    Date1        Number1
    Date2        Number2
    ...
    DateN        Number N

    I think I'm making a mistake with the Windowing because it is predicting very well with the "past days" but for the last day it ALWAYS displays 0 in any case.
    Must I add one mor row to put Date N + 1 in order to get Number N+1?
  • wpier002
    wpier002 New Altair Community Member

    Can you please explain how to perform this same process in Rapidminer 7.2.002?  When I'm searching for the window operator nothing is coming up.  Thank you.

    @jeton wrote:
    Hi!
    I'm from Argentina and new to RapidMiner so I apologize in advance for my poor English.

    I have a time series in an Excel that looks like this...

    Date              Number
    May-02        4
    May-03        9
    May-04        1
    May-05        7
    May-06        7

    and so on... Numbers (they are integers from 0 to 9) can repeat and they are "random". The last record is, for example, Sep-18 is 2. Now... I know that Sep-19 is 7 but I can't find the right model and learner to achieve that.

    Here are the steps:

    - Read Excel
    - Set Role (I labelled the date)
    - Windowing (encode series by example, horizon 1, window size 1, step size 1, create single attributes, create label (for Number))
    - Sliding Window Validation: Training (learner is SVM Evo, radial) and Testing (Apply Model, Forecasting Performance horizon 1)

    Now when I make a copy of the first 3 of the nodes and create an Apply Model node in order to "compare" (with Windowing in horizon 0) the original date with the prediction label, the last record is 0!!!! And I kwow is 3! The rest of the "predictions" are OK.

    Is there a simple way to make a prediction of the next value of an integer-valued time series based on the past values?
    Hope I've been clear enough.
    I really need your help guys!

    Thank you very much!

    Leo

    P.S.: I tried to "copy" the way of modeling from this video... http://www.youtube.com/watch?v=UmGIGEJMmN8

     

Categories