Building Models Given the Task Considering Cost, Complexity, Latency, and Scalability PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A healthcare organization is developing a model to predict patient readmission risk. Clinicians need to understand which factors (e.g., age, prior admissions, medications) drive the prediction for each patient. The team has a dataset with 50 features and is considering a random forest model. What should they do to meet the interpretability requirement?

   • AUse a deep neural network and rely on its hidden layers for interpretability.
   • BReport only the global feature importance from the random forest.
   • CUse only a linear model because random forests are inherently uninterpretable.
   • DUse SHAP values to explain individual predictions from the random forest.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   A data scientist wants to train a model using SQL queries directly on data stored in BigQuery, without moving data to a separate training environment. Which Google Cloud product is most appropriate?

   • AVertex AI AutoML
   • BVertex AI Pipelines
   • CVertex AI Workbench
   • DBigQuery ML
   Reveal Solution Discussion Select an answer first
4. 8expert · hard

   A regulatory agency needs to approve a credit scoring model. The model must be auditable, meaning every prediction must be explainable. The agency's data is in BigQuery, and they have a small team with limited ML expertise. They tested a complex gradient boosting model and a simpler logistic regression. The gradient boosting model is slightly more accurate, but the team is concerned about the effort to generate SHAP explanations at scale. What is the most appropriate approach?

   • AUse a black-box model and provide only a global explanation.
   • BUse a deep neural network and rely on its feature weights.
   • CUse gradient boosting with SHAP explanations generated in BigQuery.
   • DUse logistic regression because it is inherently interpretable and easier to audit.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A government agency needs to predict energy consumption for the next month to plan resource allocation. They have 5 years of hourly data with strong daily and weekly seasonality. The team must be able to explain the model's predictions to stakeholders. Which modeling approach is most appropriate?

   • AUse an ARIMA model with seasonal components.
   • BUse a pre-trained LLM to forecast from historical text reports.
   • CUse a random forest regressor on lagged features.
   • DUse a deep neural network with LSTM layers.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A streaming service wants to predict which movies a user will watch next. They have a large dataset of user interactions and movie metadata. They need to update the model daily and serve recommendations with low latency. The team has experience with collaborative filtering and deep learning. They are considering a matrix factorization model and a two-tower DNN. The matrix factorization is simpler and faster to train, but the DNN can incorporate side features like genre and time. What is the most appropriate approach?

   • AUse matrix factorization and serve via batch predictions daily.
   • BUse a two-tower DNN to incorporate side features and serve via an online endpoint.
   • CUse a rule-based system based on genre popularity.
   • DUse a pre-trained LLM to generate recommendations from movie descriptions.
   Reveal Solution Discussion Select an answer first