An Out-of-Bag-Probe (OOB sample) refers to a collection of data points that are not included in the training set when creating a predictive model, particularly in Ensemble-Methoden like bagging. In these methods, multiple models are trained on random subsets of the data, and the points that are not selected for a specific model’s training are considered its Out-of-Bag-Beispiele.
These OOB samples can serve a dual purpose: they provide a means of estimating the model’s performance without needing a separate validation set, and they help in evaluating the generalization ability of the model. For instance, in a Random Forest algorithm, each decision tree is built using a bootstrap sample drawn from the training data. Approximately one-third of the data points are left out of the bootstrap sample and thus are available for OOB validation.
Um zu bewerten Modellleistung using OOB samples, predictions can be made for each out-of-bag observation using the trees in the forest that did not include that observation in their training set. The aggregated results from these predictions allow for an unbiased estimate of the model’s accuracy.
Using out-of-bag samples is particularly advantageous because it maximizes data utility by leveraging the same dataset for both training and validation, thereby avoiding the pitfalls of overfitting and underfitting Wird üblicherweise mit separaten Trainings- und Validierungsdatensätzen assoziiert.