AI-ordlista
Den kompletta ordlistan över AI
Hoeffding Bound
Statistical bound used in incremental decision trees to determine with guaranteed probability whether an attribute is optimal for splitting a node, by examining a finite number of samples.
CVFDT (Concept-adapting VFDT)
Extension of VFDT incorporating concept drift detection and adaptation mechanisms, allowing dynamic modification of the tree structure without complete reconstruction.
EFDT (Extremely Fast Decision Tree)
Improved VFDT algorithm that periodically re-evaluates past splitting decisions and performs tree re-splits to correct suboptimal splits, offering better accuracy with similar complexity.
Streaming Decision Tree
Decision tree architecture specifically designed to process continuous data streams with constant memory and time constraints, without requiring complete storage of historical data.
Incremental Splitting
Node splitting process in a decision tree that occurs progressively as new data arrives, without requiring re-evaluation of the entire dataset.
Node Statistics
Aggregated counters maintained at each node of the incremental tree to track class distributions and attribute statistics, essential for split decisions based on the Hoeffding bound.
Grace Period
Minimum number of samples a node must observe before becoming eligible for splitting, ensuring statistical reliability of splitting decisions in incremental trees.
Tree Pruning (Incremental)
Adaptive pruning technique in incremental decision trees that removes branches that have become obsolete or underperforming following concept drift, continuously optimizing the tree structure.
Memory-Efficient Processing
Strategy in incremental decision trees limiting memory usage to O(log n) where n is the number of processed instances, through intelligent management of node statistics.
Pre-pruning Incremental
Anticipatory mechanism in incremental trees that prevents the creation of potentially unnecessary branches based on statistical criteria before their full development.
Attribute Evaluation
Continuous process of evaluating candidate attributes for node splitting in incremental trees, using metrics such as information gain or Gini index in an incremental manner.
Leaf Node Adaptation
Mechanism for dynamically updating predictions at tree leaves based on new instances, enabling adaptation to distribution changes without modifying the tree structure.
Tie-breaking Strategy
Tie-breaking rule used in incremental decision trees when multiple attributes have statistically equivalent performance, often based on heuristics to favor simplicity.
Instance-based Learning
Learning paradigm in incremental trees where each new instance immediately influences the model, as opposed to batch learning which requires multiple passes over the data.