Data for machine learning · Question 56 of 60
A team has only a few hundred records. Which steps are common in such cases?
3 answers are correct. Hard
Show answer
Answer
- A Use the remaining data for k-fold cross-validation
- B Train the final model with the hyperparameters found on the entire training and validation pool and evaluate it once on the holdout set
- D If feasible, set aside a small, final holdout test set
Explanation
With limited data, a three-way split can leave too little training data and increase the risk of underfitting. A small holdout test set is therefore set aside if feasible, and the rest is used for k-fold cross-validation. With the optimal hyperparameters, the final model is trained on the entire remaining pool and evaluated once on the holdout set for an unbiased assessment.