Databricks Machine Learning Associate Question 229
Select 3A data scientist is building a machine learning model to predict customer churn. They are trying to decide whether to use cross-validation or a simple train-validation split for model evaluation. Which of the following statements about cross-validation is TRUE?
- A
Cross-validation provides a more reliable estimate of model performance compared to a single train-validation split.
- B
Cross-validation is significantly faster than a train-validation split because it uses fewer data splits.
- C
Cross-validation reduces the risk of overfitting by using regularization techniques during the training process.
- D
Cross-validation can be computationally expensive, especially for large datasets or complex models.
- E
A train-validation split may produce performance estimates that are less stable than those from cross-validation.
Show answer and explanation
Correct answers: A, D, E
Explanation
Cross-validation is a useful technique for obtaining a reliable estimate of a model's generalization performance. It averages results across multiple folds, reducing dependence on a single data split. However, it is computationally intensive because the model is trained and evaluated multiple times. In contrast, a train-validation split is faster but may produce less stable and reliable performance estimates.
- A. Correct.
Correct: Cross-validation provides a more robust and reliable estimate of model performance by averaging results across multiple folds, making it less dependent on a single data split.
- B. Incorrect.
Incorrect: Cross-validation is generally slower than a train-validation split because it involves training the model multiple times on different folds of the data.
- C. Incorrect.
Incorrect: Cross-validation does not inherently apply regularization; regularization is a separate technique used during model training. Cross-validation helps evaluate the model's generalization ability.
- D. Correct.
Correct: Cross-validation involves multiple training iterations, which can be computationally expensive for large datasets or complex models.
- E. Correct.
Correct: A train-validation split can produce less stable performance estimates because the evaluation depends on a single arbitrary split of the data.