Feature Selection (via ranked feature importances and Shapely values)
Intrinsic dimension estimation
Explore SINDy (time series) style approaches for examining sparse polynomial combinations of features
H2O.ai style approaches...
Consider as a hyperparameter the threshold for the number of features to include. Where we have multiple methods for estimating feature importances, take the intersection of the top $n$ features from each method.
Consider as a hyperparameter the threshold for the number of features to include. Where we have multiple methods for estimating feature importances, take the intersection of the top $n$ features from each method.