A data science team is building a binary classification model to predict customer churn. They have a dataset with 100,000 rows and 50 features, including some highly imbalanced categorical variables. They want to use Amazon SageMaker to train a model that performs well and can be deployed for real-time inference with low latency. Which built-in algorithm would be the most appropriate starting point for this use case, considering both performance and ease of implementation?

You’ve trained a sentiment analysis model using BlazingText on a corpus of product reviews. During testing, you notice the model performs poorly on reviews containing sarcasm or industry-specific jargon. The team has decided to use a human-in-the-loop approach to improve the model by iteratively labeling the most uncertain predictions. Which SageMaker capability should you implement to operationalize this workflow efficiently?

An ML engineer is preparing to deploy a large language model for a text summarization service. The model has a size of 20 GB. The expected traffic is highly variable, with sporadic bursts, and the application requires sub-second latency. The engineer needs a deployment option that simplifies infrastructure management while meeting the performance requirements. What is the most suitable SageMaker hosting option?

Your company’s fraud detection model, deployed on a SageMaker endpoint, has been running smoothly for months. Recently, the finance team reported a gradual increase in false negatives. You suspect the underlying patterns of fraudulent transactions may have changed since the model was trained. What is the most systematic, automated approach to validate your hypothesis and decide on the next steps?

During the feature engineering phase for a recommendation model, you need to create a ‘user_engagement_score’ that aggregates ten different raw activity metrics (e.g., clicks, dwell time, downloads). This score must be calculated consistently during both training and inference. The pipeline uses SageMaker Pipelines. What is the best practice for ensuring this transformation logic is applied identically in both phases?

A team is building a pipeline to process thousands of daily satellite images, extracting features using a pre-trained computer vision model, and then running a custom clustering algorithm on those features. They want a fully managed workflow that orchestrates these steps, handles failures, and allows for easy reruns. Which combination of AWS services would provide the most cohesive and scalable solution?

You are tasked with optimizing the cost of several SageMaker training jobs that process large datasets. The jobs are not GPU-intensive and are currently using general-purpose instances. They often finish well before the scheduled hour is up, but you are billed for the full hour. Which single change is most likely to provide significant cost savings without compromising the jobs’ reliability?

After deploying a new model version to a SageMaker endpoint, you need to test its performance against the live production traffic before fully shifting over. You want to send a small, controlled percentage of user requests to the new model while sending the rest to the old, stable version. How can you achieve this safely?

An ML team stores hundreds of curated features—like customer lifetime value and average purchase size—in SageMaker Feature Store. A real-time application needs to retrieve these values for a given user ID with millisecond latency to make a prediction. How should the team configure the Feature Store to meet this latency requirement?

A data science team is building a binary classification model to predict customer churn. They have a dataset with 100,000 rows and 50 features, including some highly imbalanced categorical variables. They want to use Amazon SageMaker to train a model that performs well and can be deployed for real-time inference with low latency. Which built-in algorithm would be the most appropriate starting point for this use case, considering both performance and ease of implementation?0