Feature importances with correlated features still low even after hierarchical clustering

Question

I have a neural network, which processes correlating data. I want to output feature importances using sklearn permutation feature importance like here. Even with this, I get ver