Feature Importance

Category: science

A score that ranks which inputs are most influential in the model’s prediction.

Which variable actually matters? Is it Age, Credit Score, or ZIP code? Feature importance calculates this. It is the best tool for "Model Explainability," helping us convince stakeholders that the model isn’t using biased or illogical factors.

Common Examples

• The feature importance report showed that credit score was the dominant factor in the liability prediction, confirming our underwriting intuition.
• We utilize feature importance metrics to satisfy internal auditing and prove our models do not incorporate restricted demographic variables.