Work out feature importance share instantly with clear inputs, formula shown and shareable results.
Tree-based importances and SHAP magnitudes are only meaningful relative to the total, so a raw gain of 0.32 says nothing until you know the sum. Comparing a feature's share against the even share of 1/p exposes genuine dominance: a feature carrying 10 times the average share is doing most of the work, and is also where leakage usually hides.
Importance share
share = feature importance / total importance; even share = 1 / feature count; ratio = share / even share
Either it genuinely carries the signal, or it leaks the target. An ID, a timestamp or a post-outcome field with an extreme share should always be investigated before shipping.
They are biased toward high-cardinality features. Permutation importance or SHAP values on held-out data give a fairer ranking.