The described behavior indicates overfitting, where the model starts to memorize training data instead of generalizing.

Early stopping halts training when validation performance stops improving, preventing the model from overfitting further. AWS documentation recommends early stopping as a primary regularization technique.

Dropout randomly disables neurons during training, forcing the model to learn robust representations and reducing reliance on specific neurons. Increasing dropout is a well-established method for improving generalization.

Increasing layers or neurons increases model capacity and worsens overfitting. Model bias is unrelated to epoch-based degradation.

Therefore, Options A and B are correct.

Option D is correct because Amazon SageMaker Model Registry is the AWS feature specifically designed to catalog models, manage model versions, and organize them into model groups . AWS documentation states that you can create a Model Group and then register each model you train , and the Model Registry adds it to the Model Group as a new model version . That directly matches the requirement to automatically create versions of ML models as models are created.

The AWS docs also describe a standard workflow in which you first create a Model Group, then create an ML pipeline, and for each run of the ML pipeline , create a model version that is registered in that group. This means Model Registry supports exactly the kind of lifecycle management and version tracking needed in an MLOps workflow where new model artifacts are produced over time and need consistent governance.

The other options do not meet the requirement. Amazon ECR stores container images, not model version history. SageMaker Marketplace model packages are for discovering or subscribing to third-party or prebuilt solutions, not for automatically versioning your organization’s trained models. SageMaker ML Lineage Tracking helps trace artifacts, datasets, and processing steps for reproducibility, but AWS documentation distinguishes lineage from the actual model version management capability provided by Model Registry. Model Registry is the AWS-native feature for keeping versioned records of trained models and managing approval status, deployment readiness, and metadata in a centralized way.

Therefore, the fully verified AWS-docs answer is D , because SageMaker Model Registry is the service purpose-built to create and manage model versions as new models are produced.

Logging the metrics to Amazon CloudWatch allows the metrics to be tracked and monitored effectively.

CloudWatch Alarms can be configured to trigger when metrics breach a predefined threshold.

The alarm can be set to notify through Amazon Simple Notification Service (SNS), which can send email messages to the configured recipients.

This is the standard and most efficient way to achieve the desired functionality.

Network ACLs (Access Control Lists) operate at the subnet level and allow for rules to explicitly deny traffic from specific IP addresses. By creating an inbound rule in the network ACL to deny traffic from the suspicious IP address, the company can block traffic to the Amazon SageMaker domain from that IP. This approach works because network ACLs are evaluated before traffic reaches the security groups, making them effective for blocking traffic at the subnet level.

Step 1: An S3 event notification invokes the pipeline when new data is uploaded.

Step 2: SageMaker retrains the model by using the data in the S3 bucket.

Step 3: The pipeline deploys the model to a SageMaker endpoint.

Step 1: An S3 Event Notification Invokes the Pipeline When New Data is Uploaded

Why? The CI/CD pipeline should be triggered automatically whenever new training data is uploaded to Amazon S3. S3 event notifications can be configured to send events to AWS services like Lambda, which can then invoke AWS CodePipeline.

How? Configure the S3 bucket to send event notifications (e.g., s3:ObjectCreated:*) to AWS Lambda, which in turn triggers the CodePipeline.

Step 2: SageMaker Retrains the Model by Using the Data in the S3 Bucket

Why? The uploaded data is used to retrain the ML model to incorporate new information and maintain performance. This step is critical to updating the model with fresh data.

How? Define a SageMaker training step in the CI/CD pipeline, which reads the training data from the S3 bucket and retrains the model.

Step 3: The Pipeline Deploys the Model to a SageMaker Endpoint

Why? Once retrained, the updated model must be deployed to a SageMaker endpoint to make it available for real-time inference.

How? Add a deployment step in the CI/CD pipeline, which automates the creation or update of the SageMaker endpoint with the retrained model.

Order Summary:

An S3 event notification invokes the pipeline when new data is uploaded.

SageMaker retrains the model by using the data in the S3 bucket.

The pipeline deploys the model to a SageMaker endpoint.

This configuration ensures an automated, efficient, and scalable CI/CD pipeline for continuous retraining and deployment of the ML model in Amazon SageMaker.

Different data types require different encoding strategies. The Feedback column contains long, unstructured text responses. According to AWS ML documentation, text data must first be converted into tokens before it can be vectorized using techniques such as embeddings or bag-of-words. Tokenization is the correct preprocessing step for textual features.

The Satisfaction Level column is categorical but has a natural ordering (Low < Medium < High). AWS best practices recommend ordinal encoding for such ordered categorical variables because it preserves the inherent ranking information.

Option A is incorrect because one-hot encoding is not suitable for free-form text and would create an unmanageable number of features. Option B has the same issue for the Feedback column. Option C incorrectly applies label encoding to text and binary encoding to a three-class ordinal variable.

Therefore, tokenization for text data and ordinal encoding for satisfaction levels is the correct solution.

AWS ML best practices recommend A/B testing to validate model improvements in production while minimizing risk. By routing a controlled portion of live traffic (for example, 20%) to the new model and keeping the majority of traffic on the existing model, the company can directly compare real-world performance using the same data distribution.

This approach allows statistically meaningful comparison of business metrics such as customer retention, rather than relying solely on offline accuracy. It also limits potential negative impact if the new model underperforms in production.

Deploying the new model to 100% of traffic (Option A) introduces unnecessary risk. Offline analysis (Option C) does not reflect live user behavior. Alternating deployments (Option D) introduces confounding factors such as time-based effects.

Therefore, A/B testing is the correct solution.

For asynchronous inference on large datasets, AWS recommends SageMaker batch transform, which is designed for offline and large-scale inference workloads. Batch transform jobs do not require real-time endpoints and can process large volumes of data efficiently.

For monitoring data quality, AWS provides Amazon SageMaker Model Monitor, which supports scheduled monitoring of data quality, schema drift, and feature distribution drift. Model Monitor can publish metrics to Amazon CloudWatch and trigger alerts when thresholds are breached.

The other options lack native ML-specific monitoring capabilities. CloudTrail audits API activity, not data quality. Glue, Batch, and ECS would require extensive custom monitoring logic.

Therefore, SageMaker batch transform combined with Model Monitor is the correct solution.

Option D is correct because the company’s primary goal is to identify as many fraudulent transactions as possible . In AWS documentation, recall is defined as TP / (TP + FN), where TP is true positives and FN is false negatives. Recall measures how well a model finds all actual positive cases. In a fraud-detection setting, the “positive” class is the fraudulent transaction, so maximizing recall means minimizing missed fraud cases.

AWS documentation also explains the business tradeoff clearly: a use case that needs to correctly predict as many positive examples as possible should prioritize high recall , even if that means accepting some additional false positives and therefore only moderate precision. That matches this question exactly, because the requirement is not to be most selective or most balanced overall; it is specifically to catch the largest possible number of fraudulent transactions.

The other metrics are less aligned to the stated goal. Precision focuses on how many predicted fraud cases are actually fraud, which is most important when false positives are very costly. F1 score balances precision and recall, but the question does not ask for balance; it asks for finding as many fraudulent transactions as possible. AUC is useful for overall ranking and threshold- independent model discrimination, but it is not the most direct metric for optimizing missed-fraud detection in this scenario. Based on AWS metric definitions, when the cost of missing true positives is the main concern, recall is the best evaluation metric. Therefore, the best verified answer is D .

AWS best practices for cost governance recommend using resource tagging combined with AWS Budgets to track and alert on service-level spending. By adding tags at the SageMaker Studio user profile level, all compute resources launched by users inherit those tags automatically.

AWS Budgets supports threshold-based alerts, unlike AWS Cost Explorer, which is primarily used for historical analysis and visualization. Budgets can trigger notifications via email or Amazon SNS when spending exceeds defined limits.

IAM profiles are unrelated to cost tracking, making options C and D invalid. Therefore, tagging SageMaker user profiles and using AWS Budgets is the correct solution.