| Min. | 2nd perc. | 25th perc. | Median | 75th perc. | 90th perc. | 98th perc. | Max. |
|---|---|---|---|---|---|---|---|
| Average | Standard deviation | ||||||
| Average (clipped) | Standard deviation (clipped) |
The errors (difference between predicted and actual values) should be centered around zero, and the distribution should be "narrow", i.e the spread of the error should be limited. More generally, the errors should be "normally" distributed around zero (the curve should look like a bell).
In the error distribution histogram, errors are clipped between the 2nd and the 98th percentiles to reduce the effect of possible outliers. All values below the 2nd percentile are clipped at the 2nd percentile value ({{min}}) and all values above the 98th percentile are clipped at the 98th percentile value ({{max}}).
The average and standard deviation in the table are computed on the clipped error.
This model was evaluated using K-fold cross-test. These results are for the first fold only.