1. Why would I choose to use an Extra Trees model vs. a Random Forest model? Select the best answer.

Because it is easier to explain.

Because it doesn't add noise by resampling with replacement.

Because it improves accuracy.

Because it reduces variance and works faster.

2. I am a real estate agent building a support vector regressor to determine the price of a house I'm listing. Why should I use this model vs. an OLS model? Select the best answer.

SVR models only consider errors falling outside of a given range, creating some error tolerance.

SVR models work better for problems with high variability, and offer a more precise prediction.

SVR models work well with non-linear data.

SVR models can produce multiple outputs.

3. When would the choice between Gradient Boosting and XGBoost not matter?

When scalability is not a concern

When explainability is not a concern

When working with only continuous data points

When the accuracy score from both models is the same

4. In which of the following scenarios would you expect K-means models to struggle?

When data consists of anisotropic clusters

When data consists of circular clusters

When data is noisy

When data consists of cylindrical clusters

5. Which of the following is not a strength of hierarchical clustering?

Hierarchical clustering takes a bottom-up approach to establish clusters and can work well with non-linearly separable data

Hierarchical clustering works well with clearly defined and separable clusters

Hierarchical clustering works best with clusters of varied spreads

6. Which of the following business metrics are useful to assess the effectiveness of recommender systems?

Accuracy Score

Customer Lifetime Value

Mean Squared Error

Product Uptake

7. How does a Random Forest model compute feature importance?

By calculating the average value of gini within each split.

By calculating a weighted impurity for each node, for each feature.

By calculating the weighted impurity for each tree, for each node.

By calculating the average number of splits until nodes are pure.

8. Which of the following is a strength for stacked ensembles?

Since models are diverse, the strengths from each approach can lead to a better prediction

They process data in parallel through many models for a faster output

They give us the ability to combine classifiers and regressors in a single process

Since models are usually simplistic, the outputs are explainable and easy to compute

9. In which scenario below is soft voting better than hard voting? Select the best answer.

Predicting whether to mine in a particular geographical region

Predicting whether a credit card transaction is fraudulent

Predicting which product to recommend to a customer

Predicting whether someone has a disease

10. Collaborative filtering can use cosine similarity to find similar customers.

True

False

11. Ensemble learning is used because ________. Select the best answer.

It reduces computational inefficiencies.

It provides a better accuracy score.

It is better than other models for highly complex problems.

It reduces overfitting by layering many models together.