AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Global Feature Importance
Interpretation method that evaluates the average impact of each predictive variable on the entire model, allowing features to be prioritized according to their overall contribution to predictions.
Global SHAP Values
Game theory-based approach that quantifies the average contribution of each feature to the model's predictions across the entire dataset, ensuring mathematical consistency and additivity properties.
Partial Dependence Plot (PDP)
Visualization that shows the average marginal effect of one or two variables on the model's prediction, by marginalizing the effect of other variables to reveal global relationships.
Accumulated Local Effects (ALE)
Interpretation technique that calculates the average effect of features on local predictions, avoiding correlation biases present in PDPs and providing more reliable estimates of global effects.
Global Surrogate Models
Interpretable models (such as decision trees or linear regression) trained to mimic the global behavior of a complex black-box model, offering a simplified but understandable approximation.
Permutation Feature Importance
Agnostic method that evaluates variable importance by measuring the degradation of model performance when a feature's values are randomly permuted, revealing their global contribution.
Model-Agnostic Methods
Interpretation approaches that work with any type of machine learning model without requiring access to internal structure, relying solely on input-output relationships to analyze global behavior.
Global Feature Effects
Comprehensive analysis of each variable's impact on the model's predictions across the entire data space, combining direction, magnitude, and shape of the effect for holistic understanding.
ICE Curves (Individual Conditional Expectation)
Visualization that plots individual model predictions for different values of a feature, allowing observation of effect heterogeneity and aggregating this information for global understanding.
Friedman's H-statistic
Quantitative measure that evaluates the strength of interactions between variables in machine learning models, enabling identification of non-linear dependencies that globally affect predictions.
Global Model Visualization
Set of graphical and visual techniques that synthetically represent the global behavior of a model, including relationships between features, decision patterns, and confidence regions.
Global Feature Contribution
Quantification of the average contribution of each feature to the difference between model predictions and a reference baseline, revealing the global influence of variables on decisions.
Model-Specific Global Interpretation
Interpretation methods specifically designed for certain types of models (such as weights in neural networks or rules in decision trees) to explain their global behavior.
Global Sensitivity Analysis
Systematic study of the variation in model outputs as a function of input variations across the entire input domain, identifying the most influential factors on global behavior.
Global Rule Extraction
Process that generates a set of interpretable rules that capture the global behavior of a complex model, transforming automated predictions into explicit and generalizable knowledge.