Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Global Importance
Measure of a feature's overall influence on all model predictions, reflecting its average contribution to overall performance.
Local Importance
Evaluation of a feature's specific influence for an individual prediction, allowing understanding of the determining factors for each particular case.
Mean Decrease Impurity (MDI)
Importance metric for tree-based models calculating the average reduction in impurity (Gini or entropy) brought by each feature during splits.
Mean Decrease Accuracy (MDA)
Importance evaluation technique measuring the performance drop when a feature is removed or permuted, using out-of-bag error in random forests.
Variable Importance Plot
Graph visually representing the ranking of features in descending order of importance according to a specific metric, facilitating quick interpretation.
Model-agnostic Interpretability
Interpretation approaches working on any machine learning model without requiring access to internal structure or specific parameters.
Recursive Feature Elimination (RFE)
Feature selection technique iteratively eliminating the least important features according to a given criterion, until reaching the optimal number of variables.