DataTerminology.comNEW
Also known as: Feature relevance, Feature contribution
Techniques that score how much each input feature contributes to a model’s predictions.
DESCRIPTION
Feature importance quantifies the influence of input variables on model outputs. Tree ensembles provide built-in importances; permutation and SHAP offer model-agnostic alternatives. Importance aids model debugging, feature selection, and stakeholder communication. Each method has assumptions and biases; e.g., impurity-based scores can favor high-cardinality features. Combining multiple methods provides more robust insights. Interpret importance carefully, considering interactions and data correlations.
ALTERNATIVE DEFINITIONS
scikit-learn
Permutation importance is defined as the decrease in a model score when a single feature value is randomly shuffled.
Defines a specific, model-agnostic approach to measuring importance.
SHAP Documentation
SHAP (SHapley Additive exPlanations) assigns each feature an importance value for a particular prediction.
Highlights local attribution grounded in Shapley values.
RELATED TERMS
Partial Dependence Random Forest Gradient Boosting Explainable AI Feature Reduction Regularization LIME Global vs Local Explanations
COMPETING TERMS
SHAP Values Permutation Importance
KNOWLEDGE AREAS
TARGET AUDIENCE
Data Scientist AI Engineer MLOps Engineer Data Analyst
EXTERNAL SOURCES
- Wikipedia - Feature selection
- Interpretable Machine Learning - Comprehensive chapter on feature importance
- scikit-learn - Example: tree-based feature importances
TOP QUESTIONS
Why is Feature Importance useful?
It helps identify influential variables, guide feature selection, and build trust by explaining model behavior to stakeholders.
Category: Relevancy
How do tree-based models compute Feature Importance?
They aggregate impurity decreases or use permutation approaches. However, impurity-based scores can be biased toward high-cardinality features.
Category: Misunderstanding
What is Permutation Feature Importance?
It measures performance drop when a feature is shuffled, estimating the feature’s predictive contribution in a model-agnostic manner.
Category: Example
How do SHAP values relate to Feature Importance?
SHAP assigns additive attributions per prediction, enabling consistent local and aggregated global importance assessments.
Category: Example
What are pitfalls of Feature Importance?
Collinearity can obscure true contributions and inflate/deflate importance. Use domain context and complementary methods.
Category: Mistakes
Can Feature Importance guide feature selection?
Yes, but validate downstream performance. Removing features solely by importance can harm generalization if interactions are ignored.
Category: Requirements
How stable is Feature Importance?
Importance may vary with data splits and hyperparameters. Assess stability via bootstrapping or cross-validation.
Category: Risk
Is Feature Importance the same as causality?
No. Importance reflects predictive contribution, not causal effect. Causal inference requires additional assumptions and designs.
Category: Misunderstanding
When to prefer model-agnostic importance methods?
Use them for black-box models or when comparing across model classes. They provide consistent, comparable attributions.
Category: Alternatives