%incMSE

This is how much worse a model gets when a certain feature is randomly shuffled.

We calculate the Out of Bag MSE for the set of $B$ trees in the Random Forest. Denote this ${\text{MSE}}_{full}$.

Randomly shuffle observations.

Carry out predictions and calculate ${\text{MSE}}_{shuff}$

Determine the following quantity

$\frac{{\text{MSE}}_{shuff}-{\text{MSE}}_{full}}{{\text{MSE}}_{full}}$ Higher values of %incMSE mans the variable is more important.