Robustness of an algorithm

Hello everybody,

I'm new to knime and I have to learn and use it for train an algorithm and use it to make previsions, most probably I'll have to use a decision tree.

My problem is that I have also to demonstrate in some way the robustness of that algorithm so it can be trustable.

Is there an authomatic way inside Knime for doing such a thing? If not, do you have any suggestion?

Thanks in advance and sorry if it is this thread is in the wrong place.

Vezio

Hi,

you can use a cross validation loop. An example in the workflow in attachment.

Fabien

Hello,

if I've understood the example the way is simply to divide the data set and use a part for train the algorithm and a part for the predictor and then compare the results. Is that right?

Thanks

Vezio

You could also try bootstrap sampling.

Hi, yes you divides equally your starting sample and process many train/test. Then you look at the variability of the results for each model.

Fabien