Feature Importance Explainer Calculator
Calculate feature importance explainer with our free tool. Get data-driven results, visualizations, and actionable recommendations.
Reviewed by Daniel Agrici, Founder & Lead Developer
Formula
Importance_i = (correlation_decay^i + (1 - correlation) / n) / sum_all
Feature importance is modeled using an exponential decay weighted by the average correlation strength. Higher correlation concentrates importance in fewer features, while lower correlation distributes it more evenly. The stability score estimates reliability based on the samples-per-feature ratio.
Worked Examples
Example 1: E-commerce Purchase Prediction
Problem:A model uses 8 features to predict purchase probability. The top feature (time on site) has correlation 0.82 with the target. Dataset has 5,000 samples. What is the importance distribution?
Solution:With high correlation (0.82), importance follows a steep distribution. Top feature captures ~35% importance. Using exponential decay: Feature 1 = 35.2%, Feature 2 = 21.1%, Feature 3 = 12.7%... Cumulative top-3 explains ~69% of model variance. With 5000/8 = 625 samples per feature, stability is excellent. Minimum features needed for 95% coverage: 5 of 8.
Result:Top feature importance: 35.2% | Min features for 95% variance: 5 | Stability: 100%
Example 2: Medical Diagnosis with Many Features
Problem:A clinical model has 50 features with moderate average correlation (0.45) and 2,000 patient records. How should features be prioritized?
Solution:With moderate correlation (0.45), importance is more evenly distributed. Top feature captures ~8.5% importance. With 2000/50 = 40 samples per feature, stability is moderate (~133%). Normalized entropy is high (78.3%), indicating spread-out importance. The model likely needs 25+ features to explain 85% of variance. Consider dimensionality reduction (PCA) or L1 regularization to select a subset.
Result:Top feature: 8.5% | Min features for 85% variance: 25 | Stability: moderate
Frequently Asked Questions
What is feature importance in machine learning?
Feature importance measures how much each input variable (feature) contributes to a model prediction. It helps data scientists understand which variables drive outcomes and which can be safely removed. Common methods include permutation importance (shuffling a feature and measuring accuracy drop), Gini importance (used in tree-based models measuring impurity reduction), and SHAP values (game-theoretic approach assigning each feature a contribution). Understanding feature importance is critical for model interpretability, debugging, and building trust in AI systems.
How does permutation importance differ from Gini importance?
Permutation importance works by randomly shuffling one feature at a time and measuring the resulting drop in model performance. It is model-agnostic and gives a reliable estimate of feature relevance. Gini importance (or mean decrease in impurity) is specific to tree-based models and measures how much each feature reduces node impurity across all trees. Gini importance can be biased toward high-cardinality features, while permutation importance is generally more robust. For production models, permutation importance on a held-out test set is typically recommended.
What is the relationship between dataset size and feature importance reliability?
Larger datasets produce more stable and reliable feature importance estimates. As a rule of thumb, you need at least 30 samples per feature for basic stability, and 100+ samples per feature for robust permutation importance. With small datasets, importance rankings can be noisy and change significantly between runs. Bootstrap aggregation (computing importance across multiple data subsets) can improve stability. The stability score in Feature Importance Explainer Calculator estimates how reliable the importance rankings are given your dataset size and feature count.
References
Reviewed by Daniel Agrici, Founder & Lead Developer · Editorial policy