VotingClassifier

VotingClassifier is an ensemble meta-estimator in scikit-learn that combines the predictions of multiple base classifiers to produce a single final classification. It enables the use of diverse models, allowing them to vote on the class labels to improve predictive performance or robustness.

The estimator is configured with a list of (name, estimator) pairs and options that control how votes are tallied. The primary voting modes are hard voting and soft voting. Hard voting uses the predicted class labels from each base estimator and selects the class with the most votes. Soft voting averages the predicted class probabilities across estimators and chooses the class with the highest average probability. Weights can be assigned to individual estimators to reflect their presumed reliability, with equal weights as default. Some implementations also require all estimators to provide probability estimates when using soft voting.

During training, the VotingClassifier fits each base estimator on the same training data. For prediction, it aggregates the outputs of the base estimators according to the selected voting scheme. predict_proba is available when voting='soft' and all base estimators can provide probability estimates; otherwise, probability outputs may not be available or meaningful.

VotingClassifier is suitable for multiclass classification and can be combined with preprocessing pipelines. It is often used to stabilize predictions and combine complementary strengths of different models. Limitations include potential underperformance if the base models are poorly chosen or highly correlated, and some estimators may not provide probability estimates required for soft voting. It is not intended for multi-label problems.