KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
L1 Regularization (Lasso)
Regularization technique that penalizes the sum of absolute values of coefficients, forcing some to zero to perform automatic selection of relevant variables.
Tree-based Feature Importance
Method evaluating feature importance by their contribution to impurity reduction (Gini or entropy) in decision trees and ensembles like Random Forest.
Sequential Forward Selection (SFS)
Greedy algorithm that iteratively adds the feature that most improves model performance, starting from an empty set until reaching the desired number of variables.
Sequential Backward Selection (SBS)
Wrapper approach that sequentially removes the feature whose deletion degrades model performance the least, starting from the complete set of variables.
Correlation-based Feature Selection (CFS)
Method evaluating feature subsets based on their correlation with the class and their low inter-variable correlation to avoid redundancy.
ANOVA F-value
Statistical test calculating the ratio between between-class variance and within-class variance to evaluate the discriminant power of numerical features in classification tasks.
ReliefF Algorithm
Instance-based selection algorithm that updates feature weights based on their ability to distinguish close neighbors of the same class and different classes.