Friday, March 12, 2010

Extreme value

Arthur Charpentier shows very clearly the way that mean square error and average absolute error can diverge as outliers appear.

(I allowed negative income for simple calculations and have a nice design). We concluded that gap significantly different between the standard deviation (L2) and the absolute difference (L1) simply means that there are probably outliers in the database (outliers defined "significantly different" means " outliers ").

No comments: