I was trying to calculate manually the Root Relative Squared Error and the Relative Absolute Error given by Weka, but I can't seem to get it. Can someone show me explicitly how to do it?, Or more importantly, maybe just how to get the A bar in the denominator of the formula?
(|p1-a1|+...+|pn-an| / |a1-abar|+...+|an-abar| for Relative Absolute Error)
I attached the decision tree given out by a 10-fold cross validation in J48.
I think the output prediction might have something to do with this, since the calculation for the Mean Absolute Error and the Root Mean Squared Error had something to do with it, so I included it as well.
This link could provide some insight.
http://stackoverflow.com/questions/10776673/formula-for-relative-absolute-error-and-root-relative-squared-error-used-in
Yes, I already saw that link before I asked the question, my problem was getting the A-bar. I can't figure out how to get it. :(
Hi Jhedy,
Did you also check the ppt linked in the stackoverflow page? It seems to be well explained there. Anyway, I think the a-bar is the mean of your observations, is it not?
Cheers!
Hi! Yes, I did check the linked powerpoint in the stackoverflow page. :)
It seems I have to do the MAE and the RMSE computation twice, since you could reinterpret the equation for RAE and RRSE as MAE or RMSE divided by a second MAE or RMSE, this time using the mean of the observed values instead of the probabilities (the denominators would cancel, leaving the equations for the RAE and RRSE).
kindly visit url:
https://stackoverflow.com/questions/36876749/high-relative-absolute-error-and-root-relative-squared-error-in-classification
https://katie.mtech.edu/classes/csci347/Resources/Weka_error_measurements.pdf
- Badges
- Science topic