Knowledge check

Quiz

In Progress

Knowledge check

Evaluate your understanding of this unit by completing the knowledge check below.

Action item: Knowledge check

Work through each question. Correct answers and feedback mirror the Multiverse assessment.

1. You’re training a model on a dataset of 50,000 customer transactions. You randomly split the data into 80% training and 20% testing sets. However, your test results seem unusually high. What might have gone wrong?

A. You used too small a test set.B. The training set size is too large compared to the test set.C. You likely performed feature scaling or encoding before splitting the data.D. You didn’t shuffle the data before splitting.

2. Which of the following methods provides a more reliable performance estimate than a single train-test split, especially when the dataset is small?

A. Hold-out validationB. BootstrappingC. K-fold cross-validationD. Temporal validation

3. You’re building a sales forecasting model that relies on sequential monthly data. What validation strategy should you use to avoid data leakage?

A. Random K-fold cross-validation.B. Time-series cross-validation (rolling origin).C. Stratified sampling.D. Nested cross-validation.

4. Which of the following is not a recommended way to prevent data leakage?

A. Splitting data before preprocessing.B. Using automated ML pipelines to isolate transformations.C. Keeping test data isolated throughout model development.D. Randomly mixing test data back into training for larger sample sizes.

5. A data scientist is working on fraud detection where fraudulent cases make up only 2% of the dataset. What issue is most likely to occur if no balancing technique is applied?

A. The model will overfit the minority class.B. The model will ignore the majority class.C. The model will predict mostly the majority class and achieve misleadingly high accuracy.D. The dataset will become too large to process efficiently.

6. Which method is most appropriate when dealing with an imbalanced classification dataset with very few minority cases?

A. Undersampling the majority class.B. Oversampling the minority class using SMOTE.C. Randomly duplicating the minority class.D. Removing noisy instances from the minority class.

7. You are comparing model results before and after applying class weighting. Which metric should you use to best evaluate performance improvements on an imbalanced dataset?

A. Precision-recall or F1-score.B. AccuracyC. R-squaredD. Log-loss.

8. You discover missing values and duplicate records in your dataset. What should you do first?

A. Immediately delete all incomplete rows.B. Perform a data audit to assess the extent and pattern of missing or duplicate data.C. Impute missing values with random numbers.D. Continue model training since missing data is often insignificant.

9. A model consistently performs well on training data but fails on unseen validation data. After investigation, you find that inconsistent formatting in date fields led to duplicate rows. What best explains this issue?

A. The model lacks hyperparameter tuning.B. Data leakage between training and validation sets.C. Improper class balancing technique.D. Poor data validation and cleaning before model training.

10. You’re designing a machine learning workflow for a logistics company. Which combination of practices best ensures reliable model performance?

A. Random train-test split, accuracy as the main metric, and oversampling the majority class.B. Single hold-out validation, precision as the main metric, and no data cleaning.C. K-fold cross-validation, SMOTE for minority class balancing, and imputation for missing data.D. Stratified sampling, undersampling the minority class, and feature scaling before splitting.

In progress