Root mean square error

taghaddo

when I used the performance for my classifier  and select the root mean square error or relative error it gives me a number+/- deviations like 2+/-0.5, where 0.5 comes from? what is the confidence interval to calculate this number(for example with 90% confidence interval centered by the mean value)? please elaborate them more.

Telcontar120

If you are using cross-validation then the +/- indicates the mean performance and the standard deviation of that performance across the k cross validation folds.  However, I would be careful using this to generate confidence intervals (you could spend a long time reading many academics and all their disputes about whether this is theoretically sound or not).