ARIMA model especification tests

sebastian_gonza
sebastian_gonza New Altair Community Member
edited November 5 in Community Q&A

Hi

I am using the arima trainer but I would like to follow the optimal order of lags that should be included (I dont know if I can do this with parameter grid optimization since there are specific test like ADF etc) and I also need to perform test of normality of the residuals from an arima model, to be sure that the model is ok to forecast, and last I want to make a forecast with the variable on levels (not in the transformed state of the variable)

 

Can someone help me please?

 

Thanks 

Tagged:

Answers

  • sgenzer
    sgenzer
    Altair Employee

    hi @sebastian_gonza sorry no one has responded to this. Tagging @tftemme.


    Scott

     

  • tftemme
    tftemme New Altair Community Member

    Hi @sebastian_gonza,

     

    As you already have correctly guessed you can use the Optimize Parameters operators to find the optimal parameter settings (so, p,d,q,constant) of the ARIMA model for your own problem. The ARIMA Trainer provides you with a performance port containing the aic, bic and aicc which basically describes how well the fitted model describes the input data. Unfortunately currently there are no more performance measures (aka statistical tests) implemented. Of course specific tests differ in their results, but maybe the implemented performance measure are suitable for your problem?

    You can find a template process demonstrating this kind of auto arima in the Samples/Time Series/templates/Automized Arima on US-Consumption data in RapidMiner Studio.

     

    Another way to optimize parameters and to calculate the performance would be to fit an Arima model on a training window and calculates the forecasting capabilities on a test window. For this you can use the Forecast Validation operator. The template Samples/Time Series/templates/Forecast Validation of ARIMA Model for Lake Huron demonstrates the usage of the Forecast Validation operator. You can use the performance calculated to optimize the parameters.

     

    Concerning your last questions, I am not sure what you are asking for. The Apply Forecast operator perform the forecast for the variable in question, not for a transformed state. Could you maybe elaborate the questions a bit more.

     

    Hopes this helps,

    Best regards,
    Fabian

Categories