How to incorporate function uncertainty into machine learning algorithms?

Question

I use decision trees from Scikit. Learn to do regression in a dataset. I get very good results, but one problem that concerns me is that the relative uncertainty with respect to many functions is very high.
I tried to simply remove cases with a high degree of uncertainty, but this significantly reduces the performance of the model.

Own elements are experimentally determined; therefore, they are associated with experimental uncertainty. The data itself is not noisy.

So my question is , is there a good way to incorporate function-related uncertainty into machine learning algorithms?

Thanks for the help!

Answer 1

If undefined functions improve the algorithm, which assumes that together they are useful. However, some of them may not be. My suggestion would be to get rid of those functions that do not improve the algorithm. You can use the greedy algorithm to eliminate.

http://scikit-learn.org/stable/modules/generated/sklearn.feature_selection.RFE.html

It starts by teaching the model all the functions of the model, and then gets rid of the function that is considered the least useful. He models the model again, but with one lesser function.

Hope that helps