Glosarium AI
Kamus lengkap Kecerdasan Buatan
Out-of-Bag Score
Performance metric derived from the out-of-bag error, often expressed as 1 minus the OOB error, providing an internal evaluation of model quality without cross-validation.
OOB Estimate
Unbiased estimate of the test error obtained by aggregating predictions on out-of-bag samples for each observation in the training set.
Bagging Error
Generalization error of a bagging model, which can be efficiently estimated by the out-of-bag method without requiring an external validation set.
Random Forest OOB
Specific application of out-of-bag error to random forests, where each tree is evaluated on samples not used in its bootstrap to estimate overall performance.
OOB Variable Importance
Measure of variable importance calculated by evaluating the increase in OOB error when the values of a variable are randomly permuted in out-of-bag samples.
OOB Cross-Validation
Alternative to traditional cross-validation using out-of-bag samples as internal validation sets for each bootstrap iteration.
Bagging Variance Reduction
Fundamental property of bagging that reduces prediction variance by averaging the outputs of models trained on different bootstrap samples.
OOB Confidence Interval
Confidence interval for the generalization error estimated from the distribution of out-of-bag errors across different bootstrap samples.
Subagging OOB
Bagging variant using subsamples without replacement, where the out-of-bag estimation must be adapted to account for the different sampling strategy.
OOB Proximity Matrix
Matrix measuring the proximity between observations based on the frequency where they fall into the same terminal leaves of trees evaluated on out-of-bag samples.
Bagging Instability
Measure of the sensitivity of a base algorithm to variations in training data, a necessary condition for bagging and OOB estimation to be effective.
OOB Learning Curve
Curve showing the evolution of the out-of-bag error as a function of the number of models in the ensemble, allowing optimization of the ensemble size without overfitting.