Closed ijyliu closed 2 months ago
maybe want to just make previous rating the last step
Potentially do two stepwise sets, one with previous rating never included and one with previous rating always included
currently class weight = 'balanced' is not being picked in grid search (but we are optimizing for accuracy).
could try smote or some other oversampling/undersampling
OH 4/16/2024 options
step order - start with previous rating and show more decimal places!
consider adding ratios, on top of dollar terms versions of financial variables. shift away from just "big company" variables
Including previous rating is making the prediction problem (even for actual rating) very boring. It's hard for our NLP features to add anything on top of it.
Permutation Test - it's clearly dominating the other features
We should also ask about the order in this stepwise regression setup (note previous rating is in Categorical Feature)