Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Bootstrap Aggregating Regressor
Ensemble method that applies the bagging principle to regression models, training multiple regressors on bootstrap samples and aggregating their predictions by mean or median to reduce variance.
Out-of-Bag Error (OOB)
Bagging evaluation metric calculated on samples not included in the bootstrap sample of a given model, providing an unbiased estimate of generalization error without requiring cross-validation.
Aggregation by Mean
Prediction combination technique in bagging regressor where the final prediction is the arithmetic mean of individual predictions from all models in the ensemble.
Aggregation by Median
Robust alternative to mean aggregation where the final prediction is the median of individual predictions, less sensitive to outliers and extreme predictions from models.
Base Regressor
Individual regression algorithm (such as decision tree or k-nearest neighbors) used as a weak model in the bagging architecture, trained on different bootstrap samples.
Bagging Stability
Property of bagging regressor to produce more stable and less variable predictions in response to changes in training data, thanks to variance reduction through aggregation.
Random Subspace Regressor
Bagging variant where each base regressor is trained on a random subset of features in addition to bootstrap samples, increasing model diversity.
Pasting Regressor
Bagging variant where samples for training each model are drawn without replacement, used when the dataset is too large for bootstrap with replacement.
Aggregated Mean Squared Error
Evaluation metric specific to bagging regressors calculating the average of squared errors between aggregated predictions and true values on the test set.
Bagging with Decision Trees
Most common application of bagging regressor using decision trees as base models, known for significantly reducing variance while maintaining low bias.
Bagging Prediction Intervals
Technique using the distribution of individual predictions from bagging models to construct quantitative confidence intervals around the final aggregated prediction.