AI 용어집
인공지능 완전 사전
Temporal Cross-Validation
Evaluation method specific to temporal data that respects chronological order, using past data to predict future data. Essential to avoid temporal information leakage in time series and forecasting.
Group Cross-Validation
Technique that ensures observations from the same group never appear in both training and test partitions simultaneously. Essential when data has intra-group correlation structures like repeated measures or hierarchical data.
AUC (Area Under Curve)
Aggregate performance measure calculated as the area under the ROC curve, ranging from 0 to 1 where 1 represents perfect classification. Robust indicator of model discriminative ability independent of classification threshold.
Brier Score
Evaluation metric for predictive probabilities calculated as the mean squared differences between predicted probabilities and observed outcomes. Particularly useful for assessing the calibration of probabilistic models rather than just their discrimination.
Monte Carlo Cross-Validation
Iterative evaluation method that randomly creates training and test partitions at each iteration, allowing estimates with controlled overlaps. Offers flexibility in partition sizes but may introduce biases.