I am currently trying to build random forest models to infer the relevance of a list of features in 70 subjects. Currently, I have 460 features, I've read that models should not be built using more features than samples. I understand that feature selection can be applied to filter more meaningful features.
I wonder if anyone could help me out and explain what happens to a random forest model when it is "overly" trained. And is there an optimal row x columns ratio? I mean for N samples should models be built using N/2 features or square root of N?
Do I have to calibrate my model to find this optimal ratio? How to identify it?
Thanks in advance.
question from:https://stackoverflow.com/questions/66062932/samples-x-features-ratio-in-rf-models
