For a very large dataset (e.g., millions of records), which training/validation/test split ratio is most appropriate to balance model performance with computational efficiency?

98% training, 1% validation, 1% test

50% training, 25% validation, 25% test

70% training, 20% validation, 10% test

90% training, 5% validation, 5% test

A data scientist is preparing a dataset for a recommendation system. They have user-item interaction data. They plan to split the data randomly into training and test sets. What is the most likely problem with this approach?

The model will not be able to learn from the data.

The class distribution will be imbalanced.

Users in the test set may also appear in the training set, causing the model to memorize user preferences.

The test set will be too small.

To prevent data leakage, when should feature selection or hyperparameter tuning be performed?

On the entire dataset before splitting

Only on the training set, using cross-validation or a validation set

A data scientist is preparing a dataset for a churn prediction model. They perform feature scaling (e.g., standardization) on the entire dataset before splitting into training and test sets. What is the primary problem with this approach?

It reduces the size of the training set, hurting model performance.

It makes the model unable to handle new data at inference time.

It changes the class distribution in the training and test sets.

It causes the test set to influence the training process, leading to data leakage.

What is the main benefit of using stratified splitting instead of simple random splitting for a classification problem with a rare class?

It guarantees the model will have high accuracy on the rare class

It eliminates the need for a validation set

It increases the total number of samples in the dataset

It ensures the rare class appears in the training and test sets in proportions similar to the original dataset

CertNexusCertified Data Science Practitioner (CDSP)

Domain 4Objective 1

Objective 4.1 Prepare Data Sets for Modeling CERTIFIED-DATA-SCIENCE-PRACTITIONER Practice Questions (Page 3)

Part of the 4.0 Building models domain, which accounts for 19-27% of the CERTIFIED-DATA-SCIENCE-PRACTITIONER exam.

26questions here

6free pages

7concepts

19-27%of the exam

Questions 11–15

11foundation · easy
For a very large dataset (e.g., millions of records), which training/validation/test split ratio is most appropriate to balance model performance with computational efficiency?
Select an answer first
12expert · hard
A data scientist is preparing a dataset for a recommendation system. They have user-item interaction data. They plan to split the data randomly into training and test sets. What is the most likely problem with this approach?
Select an answer first
13foundation · easy
To prevent data leakage, when should feature selection or hyperparameter tuning be performed?
Select an answer first
14application · medium
A data scientist is preparing a dataset for a churn prediction model. They perform feature scaling (e.g., standardization) on the entire dataset before splitting into training and test sets. What is the primary problem with this approach?
Select an answer first
15foundation · easy
What is the main benefit of using stratified splitting instead of simple random splitting for a classification problem with a rare class?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to CERTIFIED-DATA-SCIENCE-PRACTITIONER

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by CertNexus. “CERTIFIED-DATA-SCIENCE-PRACTITIONER” is a trademark of its owner, used for identification only.