Databricks Machine Learning Associate Question 584
Select 3A data scientist is working on a machine learning project and needs to evaluate the generalization performance of their model. They are deciding between using cross-validation and a simple train-validation split. Which of the following are valid considerations when choosing cross-validation over a train-validation split?
- A
Cross-validation provides a more robust estimate of model performance by averaging results across multiple splits.
- B
Cross-validation tends to be computationally more expensive than a single train-validation split.
- C
Cross-validation guarantees that the model will perform better on unseen data compared to a train-validation split.
- D
Cross-validation helps reduce the effect of variability that may result from a single random train-validation split.
- E
A train-validation split is always preferred for large datasets because cross-validation cannot handle them.
Show answer and explanation
Correct answers: A, B, D
Explanation
Cross-validation is a robust method for estimating model performance as it reduces the impact of variability from a single random split and provides a more reliable estimate by averaging results across multiple splits. However, it is computationally expensive compared to a single train-validation split. While cross-validation does not guarantee better performance on unseen data, it is a preferred approach when computational resources and time allow, especially for smaller datasets or when a more reliable performance estimate is needed.
- A. Correct.
Correct: Cross-validation evaluates the model on multiple splits, reducing the likelihood of overfitting to a single split and providing a more reliable estimate of performance.
- B. Correct.
Correct: Since cross-validation involves training and validating the model multiple times (once for each split), it is more computationally intensive compared to performing a single train-validation split.
- C. Incorrect.
Incorrect: Cross-validation does not guarantee better performance on unseen data. It only provides a more reliable estimate of how well the model might generalize.
- D. Correct.
Correct: By averaging results across multiple splits, cross-validation reduces the variability introduced by a single random train-validation split.
- E. Incorrect.
Incorrect: Cross-validation can handle large datasets, though it might require more computational resources. It is not always true that a train-validation split is preferred for large datasets.