1500 Questions | AWS Machine Learning Specialty 2026

Domain 2: Machine Learning Implementation (21%): Selecting optimal algorithms (Linear Learner, XGBoost, DeepAR, etc.), hyperparameter tuning, and advanced model evaluation.

Domain 3: Model Operations and Deployment (20%): Deploying models via SageMaker endpoints, implementing A/B testing, and managing model versioning/rollbacks.

Domain 4: Data and Analytics (21%): Implementing data visualization strategies, reporting, and maintaining strict data governance and management.

Domain 5: Machine Learning Engineering and Operations (18%): Designing scalable ML architectures, ensuring model security/compliance, and optimizing for cost and performance.

B. Increase the instance count of the training cluster by 50%.

C. Change the file format from CSV to Parquet without modifying the data.

D. Use Amazon S3 Select to filter out the majority class randomly.

E. Disable hyperparameter tuning to save time on imbalanced data.

F. Manually delete 90% of the minority class data points.

Correct Answer: A

Explanation:

B (Incorrect): Scaling hardware does not fix the underlying statistical distribution of the data.

C (Incorrect): While Parquet is more efficient for storage, it does not address class imbalance.

D (Incorrect): S3 Select is a retrieval tool; random filtering without a strategy can lead to loss of valuable information.

E (Incorrect): Tuning is still necessary, though the evaluation metric (like F1-score) becomes more important than accuracy.

F (Incorrect): Deleting minority data makes the problem worse; you want more data from the minority class, not less.

Question 2: Which AWS service is best suited for real-time data ingestion and transformation of streaming data before it is stored in an S3 bucket for ML training?

B. Amazon Kinesis Data Firehose

C. AWS Snowball Edge

D. Amazon RDS Read Replicas

E. Amazon S3 Glacier Deep Archive

F. AWS Batch

Correct Answer: B

Explanation:

A (Incorrect): DataBrew is a visual data preparation tool, primarily used for cleaning static datasets, not real-time streaming ingestion.

C (Incorrect): Snowball is for offline, physical data migration of petabyte-scale data.

D (Incorrect): Read replicas are for scaling database read traffic, not ingestion of external streams.

E (Incorrect): Glacier is for long-term archival, not active ingestion or processing.

F (Incorrect): AWS Batch is for batch computing jobs, not real-time stream processing.

Question 3: A Machine Learning Engineer wants to deploy a model that requires low latency but also wants to minimize costs by only paying when the model is actually being invoked. Which SageMaker deployment option should I choose?

B. SageMaker Batch Transform.

C. SageMaker Serverless Inference.

D. SageMaker Canvas.

E. Amazon Elastic Inference.

F. SageMaker Multi-Model Endpoints with dedicated instances.

Correct Answer: C

Explanation:

A (Incorrect): Real-time inference with P3 instances involves a constant hourly cost for the running instances.

B (Incorrect): Batch Transform is for processing large datasets at once, not for real-time low-latency requests.

D (Incorrect): Canvas is a no-code visual interface for building models, not a deployment type.

E (Incorrect): This is an acceleration tool for instances, not a standalone serverless deployment option.

F (Incorrect): Multi-model endpoints still require a persistent underlying instance that you pay for hourly.

You can retake the exams as many times as you want

This is a huge original question bank

You get support from instructors if you have questions

Each question has a detailed explanation

Mobile-compatible with the Udemy app

30-days money-back guarantee if you're not satisfied