[Biomod-commits] Variable importance

[wilfried.thuiller at ujf-grenoble.fr]

Dear Jesus,
This is not implemented in BIOMOD.
But this is rather easy to compute.

You load the specific model you look at.
You make prediction on your original data
You make prediction on your original data but after having  
re-shuffling explanatory variable 1
Then look at the correlation between the prediction and the prediction  
with one variable randomized.
Repeat it 100 (or much less if this is time consuming).
Compute the mean correlation (or the median)
Then you have your measure of variable importance. If the correlation  
is high, it means the variable has no importance (because it has been  
randomized). If this is the way around, strong variable.

You repeat the process for each variable.
Then, for each repetition, and for each model.

Does it help?

Wilfried



Jesus <jeaggu at gmail.com> a écrit :

> Hello all,
>
> I constructed SDM with 5 different algorithms for different species doing
> 10 runs per algorithm per species and now I would like to check the
> variables importance. I so far know that I can get the variable importance
> results only for the final model, do you know if there is a way in order to
> get the results also for each run?
>
> Cheers
>