AI-ordlista
Den kompletta ordlistan över AI
CART Pruning
Process of reducing tree complexity by eliminating the least significant branches to avoid overfitting while minimizing validation error.
Binary Splitting
Specific method in CART where each parent node is split into exactly two child nodes using the optimal splitting criterion.
Complexity Parameter
Alpha coefficient in CART pruning that penalizes tree size, balancing model accuracy and simplicity.
Maximal Tree
Fully developed decision tree without pruning, where each leaf contains observations of a single class or meets a predefined stopping criterion.
Minimum Impurity
Predefined threshold in CART that stops splitting a node when impurity (Gini or variance) is below this value.
Pruning Sequence
Ordered set of trees with decreasing complexity generated by CART, each being a pruned version of the previous one.
Resubstitution Error
Error rate calculated on training data used by CART as a reference to evaluate the impact of each potential pruning.
CART Optimization
Greedy algorithm that performs locally optimal splits at each node without guaranteeing global optimality of the final tree.
CART Stability
Sensitivity of the CART algorithm to variations in the training data, which can lead to significantly different tree structures.
Handling Missing Values
Specific CART approach that redirects observations with missing values to the most appropriate child node based on conditional probabilities.
Class Weighting
Technique in CART to handle class imbalances by adjusting the weights in the calculation of the Gini index.
V-fold Cross-Validation
Method used with CART to evaluate the performance of different pruning sequences and select the optimal tree.
CART Decision Rule
Set of if-then logical conditions extracted from the path from the root to a leaf, interpretable and directly applicable for prediction.