What do you think about the idea of creating a random forest using DL for tabular data?
There seems to be no reason this is not possible. There’s a reason I think it’s good not only in concept but also in performance. It is not a good performance that is required of the models that consist of RF, but RF requires the models to be overfitting as possible and to be different from each other as possible. This seems to be possible enough for DL for tabular data.
The nice thing about this is that the model could deal with recursion on tabular data. DL is weak at processing tabular data and tree-based models cannot handle recursion. So it would be nice to be good at both.
I looked for a while but couldn’t find anything about this. I wonder if there’s anything I couldn’t find…