Databricks Machine Learning Associate Question 583
Select 3You are working on a machine learning project in Databricks and need to evaluate the performance of your model. You are deciding between using cross-validation and a simple train-validation split. Which of the following statements correctly describe the benefits and downsides of using cross-validation over a train-validation split?
- A
Cross-validation provides a more robust estimate of model performance by averaging results across multiple folds.
- B
Cross-validation is computationally more expensive compared to a train-validation split.
- C
Cross-validation completely eliminates the risk of overfitting the model to the training data.
- D
A train-validation split is faster to execute but may provide a less stable estimate of model performance if the dataset is small.
- E
Cross-validation is the preferred approach when the dataset is very large and computational resources are limited.
Show answer and explanation
Correct answers: A, B, D
Explanation
Cross-validation is a robust technique for evaluating model performance as it averages results across multiple folds, reducing the risk of biased estimates. However, it is computationally expensive and not always suitable for very large datasets or limited resources. A train-validation split is faster but may lead to less reliable performance estimates, particularly for small datasets.
- A. Correct.
Correct: Cross-validation uses multiple folds to evaluate the model, reducing the likelihood of performance fluctuations caused by a single train-test split.
- B. Correct.
Correct: Since cross-validation involves training and evaluating the model multiple times (once for each fold), it is computationally more demanding than a simple train-validation split.
- C. Incorrect.
Incorrect: While cross-validation reduces the likelihood of overfitting, it does not completely eliminate it. Overfitting can still occur during model training.
- D. Correct.
Correct: A train-validation split is faster to perform but may lead to less reliable performance estimates, especially with small datasets where the train-test division might not be representative.
- E. Incorrect.
Incorrect: Cross-validation is generally not the preferred approach for very large datasets with limited computational resources. In such cases, a train-validation split is often chosen due to its efficiency.