AWS Certified Machine Learning Engineer - Associate Practice Test
The AWS Certified Machine Learning Engineer - Associate (MLA-C02) concerns implementing, deploying, and maintaining AI and machine learning solutions on AWS. The work connects data preparation and model training with secure access to cloud resources and production inference. Practice topics include selecting evaluation metrics, preparing features, configuring SageMaker jobs, and choosing managed deployment options for different traffic patterns.
Exam overview
About the AWS Certified Machine Learning Engineer - Associate Exam
Machine learning engineering requires decisions across the model lifecycle. For imbalanced classification, precision, recall, and related metrics can reveal performance that accuracy conceals. Data workflows involve transforming tabular features, handling missing values, and passing S3 input and artifact locations to custom training containers. AWS scenarios also test least-privilege access through IAM, private-subnet networking, and container image storage. Deployment choices depend on latency and traffic: managed SageMaker endpoints support low-latency inference, while serverless inference can suit intermittent requests. Operational workflows may connect reusable processing steps in SageMaker Pipelines with model testing, approval states, and human review. These tasks require distinguishing services by their role and reasoning about how security, data movement, training, and deployment decisions affect a working ML solution.
Sample Practice Questions
Review a selection from this exam's published US Practice Tests practice bank.
A dataset is used to train a binary classifier, and the class distribution is highly imbalanced. Which evaluation metric is generally more informative than accuracy in this situation?
Show answer
Correct answer: APrecision-Recall AUC is often more useful than accuracy for imbalanced classification because it focuses on the tradeoff between precision and recall. Accuracy can be misleading when one class dominates.
Which AWS service would you most likely use to control who can start training jobs, deploy endpoints, and access related AWS resources in a least-privilege manner?
Show answer
Correct answer: AIAM is the AWS service used to define permissions and enforce least-privilege access to AWS resources, including ML-related actions. CloudFront, ElastiCache, and Direct Connect do not manage access permissions.
A team wants to run a custom training container on SageMaker, and the container must read input data from Amazon S3 and write model artifacts back to S3. How does SageMaker typically provide these locations to the container?
Show answer
Correct answer: ASageMaker passes input and output locations to training containers using environment variables and standard local filesystem paths that map to S3 data channels and output directories. This is the normal integration pattern for custom containers.
A team wants to build a reusable data processing step in AWS that can be run as part of an ML pipeline to transform tabular data before training. Which SageMaker capability is best suited for this purpose?
Show answer
Correct answer: ASageMaker Processing runs containerized processing jobs for data preparation, feature engineering, evaluation, and other batch transformations. It is a good fit for reusable preprocessing steps in an ML workflow.
Which AWS service is commonly used to store container images for SageMaker training or inference jobs?
Show answer
Correct answer: AAmazon Elastic Container Registry (ECR) is the managed registry for Docker container images and is commonly used with SageMaker custom containers.
You have a tabular dataset in Amazon S3 and want to perform feature engineering such as normalizing numeric columns, encoding categorical values, and handling missing data before training. Which managed AWS service is specifically built for large-scale data preparation and feature engineering?
Show answer
Correct answer: ASageMaker Data Wrangler is purpose-built for preparing data and creating features for machine learning workflows. It supports common transformations used before training and can export results into SageMaker workflows.
A production model receives requests with highly variable traffic, including long idle periods. You want an inference option that can scale down quickly and avoid paying for continuously provisioned instances when there is no traffic. Which option is most appropriate?
Show answer
Correct answer: AServerless inference is appropriate for intermittent or spiky request patterns because it automatically scales and does not require continuously running endpoints.
How to use this practice test
The free practice bank contains 50 published questions; each session presents up to 20 random questions. Complete a session to see your results, then revisit missed concepts such as metric selection, SageMaker data inputs, IAM permissions, feature preparation, and inference options. Use each result to identify which AWS service or ML decision needs further review.
Preparation tips
- For imbalanced binary classification scenarios, compare accuracy with precision, recall, F1 score, and precision-recall curves; identify which error type matters most.
- Trace a custom SageMaker training job from its IAM execution role to S3 input data, output artifacts, and the container image stored in Amazon ECR.
- Practice mapping tabular-data tasks—missing-value handling, normalization, and categorical encoding—to suitable AWS data-preparation and SageMaker pipeline capabilities.
- Compare SageMaker real-time and serverless inference by latency needs and traffic patterns, including whether long idle periods make scale-to-zero behavior useful.
- Review how VPC private subnets restrict network paths for training jobs, and distinguish automated model checks from approval workflows that require human review.