AWS Certified Machine Learning - Specialty practice questions.

An ML team needs to land 10 GB/hour of clickstream events into S3 in near real time, partitioned by event date. They want the lowest operational overhead. Which combination best fits?

A classification model trained on a 95/5 imbalanced binary target shows 95% accuracy on the test set. Which interpretation is correct?

A team needs to deploy a model that receives infrequent inference requests with strict cost constraints; they accept occasional cold-start latency. Which SageMaker deployment option fits?

A team must detect data drift on a production model endpoint and trigger alarms when the distribution of one input feature shifts beyond a defined threshold. Which AWS feature is built for this?

A business wants to extract structured data (key-value pairs, tables, signatures) from PDF tax forms with minimal custom training. Which AWS service is the best fit?

A team needs to compare three foundation models on a private evaluation dataset before choosing one for a production RAG application. Which AWS Bedrock feature is designed for this comparison?

A team is preparing features that will be reused across multiple training jobs and online inference endpoints. They want consistent features across training and serving with point-in-time correctness. Which AWS service fits?