Absolute error?

masquerade23masquerade23 Member Posts: 2 Learner I
I really need some help regarding the absolute error. After I have done the Performance(Regression) I get these values back as absolute error 1.247 +/- 0.369. Does this means the error is between +1.247 and -0.369? I really have no idea how to interpret these value.

Answers

  • varunm1varunm1 Member Posts: 661   Unicorn
    edited June 5
    Hello @Patrick_Haas

    The error is between (1.247+0.369) or (1.247-0.369). 

    @mschmitz any suggestion on this. This calculation with +/- is based on Confidence interval in statistics right?

    Regards,
    Varun
  • masquerade23masquerade23 Member Posts: 2 Learner I
    Hello @varunm1

    Thank you so much for your help!

    Regards,
    Patrick
    varunm1
  • mschmitzmschmitz Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 2,074  RM Data Scientist
    @masquerade23,
    to add to this, this gives you the amount of variance in the different folds of a validation. So in the different folds you havesometimes values like 1.2 but sometimes also an error of 0.9 or 3.0

    Best,
    Martin
    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
    varunm1sgenzer
  • Telcontar120Telcontar120 Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,210   Unicorn
    The plus/minus component is just the standard deviation around the reported average of the performance across the different cross validation folds.  So it is related to confidence intervals but is not directly computed or dependent on them.
    Brian T.
    Lindon Ventures 
    Data Science Consulting from Certified RapidMiner Experts
    varunm1sgenzer
  • jacobcybulskijacobcybulski Member, University Professor Posts: 83   Unicorn
    Remember that in cross-validation, each fold creates a different sample with its own regression line, and each regression model produces a slightly different set of estimates and errors. The MAE is calculated for each and their mean is what RM returns, indicating a range of those errors. Why is it significant? This is because a regression models for the entire population (and the data yet to be collected) falls in between the regression models you produce in cross-validation (which define a confidence interval). In this way, RapidMiner gives you an indication of the estimated MAE for the population and the future data.
    sgenzer
Sign In or Register to comment.