What is the default setting in AutoML for the data split in model evaluation?

The default data split setting in AutoML for model evaluation is structured as 80% of the data for training, 10% for validation, and 10% for testing. This approach allows the model to learn from a significant portion of the dataset while still setting aside a sufficient amount to tune the parameters during validation and evaluate the final performance on unseen data.

The rationale for this 80-10-10 split is to ensure that the model is well-trained with enough data, as adequate training data helps enhance the model's ability to generalize to new, unseen examples. The validation set serves for hyperparameter tuning and selecting the best model configuration without biasing the evaluation metric through exposure to the test set, which should remain untainted until the final evaluation.

This balanced distribution optimizes for both effective model training and robust evaluation, helping in reducing overfitting while still providing reliable metrics when assessing performance.