🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉

Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.

CLICK HERE TO DOWNLOAD

¡Cuál es el porcentaje de error relativo que me indica que mi modelo es válido?

MiguelHH98MiguelHH98 Member Posts: 11 Contributor I
edited March 2020 in Help
Hola!

Mi modelo incluye aproximadamente 100 datos por cada variable. El error relativo resultante al correr el modelo es de 7.9% bajo el método elegido como "best performance". Quisiera saber si con ese error mi modelo es válido o, si es que no, cuál sería el adecuado. 

Espero su pronta respuesta. Gracias.

Saludos,

Miguel Hinostroza
Jasmine_

Best Answers

  • varunm1varunm1 Moderator, Member Posts: 1,207   Unicorn
    edited March 2020 Solution Accepted
    Hello @MiguelHH98

    Based on the results, the GBT has better performance than others. So, to define if the model is valid, my choice is to look at correlation values and RMSE. You can take R^2 (squared correlation) as a reference. An R^2 >0.5 is good for many domains. In your case, you can select the option correlation from the drop-down menu (instead of relative error) and calculate the square of correlation to see if you have good R^2 value. 

    The standard values change based on domains and the amount of error acceptance is also based on domain knowledge.
    Regards,
    Varun
    https://www.varunmandalapu.com/

    Be Safe. Follow precautions and Maintain Social Distancing

    Jasmine_MiguelHH98

Answers

  • [Deleted User][Deleted User] Posts: 0 Learner III
    edited March 2020
    Hello
    @MiguelHH98

    Please show the result of your process ( with screen shot) to see that and can explain your result. :)

    Regards
    mbs
    Jasmine_
  • MiguelHH98MiguelHH98 Member Posts: 11 Contributor I
    Hi, mbs:

    These are the results:


    I used, approximately, 20 attributes and 100 values per attribut. I want to know if my model is valid according to the best relative error (Gradient Boosted Trees) or what percentage could you recommend. Thanks.

    Regards,

    MiguelHH98
    Jasmine_
  • MiguelHH98MiguelHH98 Member Posts: 11 Contributor I
    I appreciate the advices. Thank you, @mbs and @varunm1
    [Deleted User]sgenzerJasmine_
Sign In or Register to comment.