Building Models Given the Task Considering Cost, Complexity, Latency, and Scalability PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A bank is developing a model to detect fraudulent transactions. They need to explain to customers why a transaction was flagged. The bank has a large dataset and a team of ML engineers. They tested a random forest and a deep neural network. The random forest has an AUC of 0.92, and the DNN has an AUC of 0.93. The bank's compliance team requires that explanations be consistent and reliable. What is the most appropriate approach?

   • AUse the random forest with SHAP explanations.
   • BUse a linear model because it is the most interpretable.
   • CUse the DNN and provide only global feature importance.
   • DUse the DNN with LIME explanations.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   A team needs to forecast monthly sales for a retail chain. The data is a single time series with clear seasonality, and the team requires low latency and minimal infrastructure cost. Which model type is most appropriate for this task?

   • AA deep neural network (DNN) with multiple hidden layers
   • BAn ARIMA model with seasonal components
   • CA large language model (LLM) fine-tuned on sales data
   • DA gradient-boosted tree model with lag features
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   A team needs to build a custom ML model with full control over the training code, hyperparameters, and infrastructure. They also want to automate the training pipeline. Which Google Cloud product is most appropriate?

   • ABigQuery ML
   • BVertex AI Feature Store
   • CVertex AI AutoML
   • DVertex AI Pipelines
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   An ML engineer must choose a model for real-time fraud detection on streaming transactions. The model must score each transaction in under 10 milliseconds and handle high throughput. Which model type is most appropriate?

   • AA large language model (LLM) with few-shot prompting
   • BA gradient-boosted tree model
   • CAn ARIMA model
   • DA deep neural network (DNN) with several layers
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A financial services company needs to approve credit card transactions in real time. The model must return a decision in under 100 milliseconds to avoid customer friction. The system processes thousands of transactions per second, and the model is a gradient-boosted tree. Which deployment strategy is most appropriate?

   • AOnline prediction with a Vertex AI endpoint using autoscaling.
   • BUse BigQuery ML to run predictions on the transaction stream.
   • CBatch prediction on a scheduled basis every few minutes.
   • DEdge deployment on each merchant's point-of-sale device.
   Reveal Solution Discussion Select an answer first