AI 용어집
인공지능 완전 사전
Group Cross-Validation
Cross-validation technique where observations are grouped according to predefined criteria to prevent information leakage between training and test sets.
Leave-One-Group-Out (LOGO)
Cross-validation variant where an entire group is left out for testing at each iteration, ensuring complete separation of grouped data.
Stratified Group K-Fold
Combination of stratified K-Fold and group K-Fold preserving both class distribution and group integrity in each partition.
Group Shuffle Split
Cross-validation technique randomly distributing groups between training and test sets with control over the number of iterations and proportions.
Time Series Group Split
Cross-validation adapted for grouped time series data respecting chronological order while preventing leakage between temporally correlated groups.
Nested Group Cross-Validation
Two-level cross-validation using groups to prevent overfitting during hyperparameter selection and final model evaluation.
Group-aware Feature Selection
Feature selection process considering group structure to avoid selecting features that introduce information leakage.
Group Leakage
Phenomenon where information from a group appears in both training and test sets, artificially biasing model performance.
Group-wise Scoring
Evaluation method calculating performance metrics by group before aggregating them, allowing identification of performance disparities between groups.
Hierarchical Group Cross-Validation
Advanced technique handling nested or hierarchical group structures to preserve multi-level dependency relationships.
Group Blocking
Strategy explicitly preventing observations from the same group from being split between training and test sets during cross-validation.
Group-based Bootstrapping
Resampling method where entire groups are drawn with replacement rather than individual observations, preserving dependency structure.
Multi-level Group Cross-Validation
Cross-validation simultaneously handling multiple grouping levels for complex data structures with cross-dependencies.
Group-aware Hyperparameter Tuning
Hyperparameter optimization using group cross-validation to ensure unbiased evaluation of model performance.
Group Imbalance Handling
Adaptive techniques for handling imbalances in group size or representation during cross-validation.
Cross-Group Generalization
Model's ability to perform on groups not seen during training, specifically evaluated through group cross-validation.
Group-aware Pipeline
Processing chain integrating group management at each stage, from preprocessing to final evaluation, including training.