Description
For learners studying for AWS Certified Machine Learning Engineer - Associate (MLA-C02)
What you'll learn
Practice pack content
Domain 1 - Data Preparation for ML and AI
Domain 2 – ML Model and Foundation Model (FM) Development
Domain 3 – Deployment and Orchestration of ML and AI Workflows
Domain 4 – Operating, Monitoring, and Securing ML and AI Solutions
Final Exam Simulators
Exam Study Guide
Interactive Flashcards
Try 3 Machine Learning Engineer - Associate (MLA-C02) practice questions
Choose an answer, explore the explanation, and see why the other options are less suitable. Three short scenarios. No sign-in required.
1. Correcting a heavily imbalanced fraud dataset
A fraud-detection training dataset contains 99.5% legitimate transactions and 0.5% fraudulent transactions. A classifier reports very high overall accuracy but misses many fraudulent cases. Which data-preparation action most directly addresses the underlying training-data problem?
Answer and full explanation
Correct answer: B — Rebalance the classes before retraining.
Why B is correct
The decisive issue is class imbalance: fraudulent transactions are severely underrepresented compared with legitimate transactions. MLA-C02 Task 1.3 explicitly includes resolving class imbalance as part of validating data quality and managing bias. AWS SageMaker Data Wrangler supports balancing operations such as random oversampling, random undersampling, and SMOTE for underrepresented categories. Rebalancing gives the model more useful representation of the minority class and can improve learning for fraud cases. Afterward, the team should also evaluate metrics such as precision, recall, and F1 rather than relying only on overall accuracy.
Why the other options are less suitable
- A — Normalize features only: Feature scaling can be important for some algorithms, but it does not change the severe ratio between legitimate and fraudulent examples. The minority class would remain underrepresented.
- C — Remove fraudulent records: Removing minority-class examples makes the imbalance worse and gives the model even less evidence about fraud patterns. The objective is to improve representation, not reduce it.
- D — Train for more epochs: Additional epochs do not correct an unbalanced class distribution. They can cause the model to fit the majority pattern more strongly without improving minority-class detection.
Key takeaway: When one target class is severely underrepresented, address the class distribution before trusting aggregate accuracy.
Read two more sample questions
2. Running nightly inference without a persistent endpoint
A company must score 20 million records stored in Amazon S3 once each night. No application requires an immediate prediction, and the results can be delivered several hours later. Which SageMaker AI inference option best matches the workload?
Answer and full explanation
Correct answer: C — Use SageMaker AI batch transform.
Why C is correct
Batch transform is intended for inference over large datasets when an always-on endpoint is not required. The workload is scheduled, processes millions of stored records, and does not need an immediate response, which are strong indicators for batch inference. MLA-C02 Task 3.1 explicitly requires selecting model inference strategies such as real-time and batch processing. AWS documentation states that SageMaker AI batch transform can produce inferences for large datasets and can be used when a persistent endpoint is unnecessary. This approach aligns the deployment model with both the latency and workload pattern in the scenario.
Why the other options are less suitable
- A — Real-time endpoint: A real-time endpoint is appropriate when low-latency online predictions are needed. Keeping capacity running continuously would add unnecessary cost for a once-nightly batch workload.
- B — Serverless inference endpoint: Serverless inference can help with intermittent online requests, but the requirement is a large scheduled dataset rather than request-driven interactive inference. Batch transform is more directly aligned to the workload.
- D — Asynchronous endpoint: Asynchronous inference is useful for queued requests with longer processing times, especially when clients submit individual inference requests. The scenario already has one large nightly dataset that can be processed as a batch job.
Key takeaway: Choose batch inference when large stored datasets can be processed offline and no persistent endpoint is needed.
3. Detecting production data drift
A production model has become less accurate after customer behavior changed. The ML team wants an automated way to detect when incoming feature distributions differ materially from the training baseline and alert operators. Which approach best addresses the requirement?
Answer and full explanation
Correct answer: A — Monitor production distributions against a baseline and alert on drift.
Why A is correct
Data drift occurs when the statistical characteristics of production inputs move away from the data distribution used to train or baseline a model. MLA-C02 Task 4.1 explicitly includes detecting changes in data distribution and using drift-detection pipelines to monitor production performance. AWS monitoring guidance uses baseline statistics and constraints as a reference and compares current inference data against that baseline to identify violations. Automated scheduling and alerting make it possible to detect the change before degradation becomes severe. This is the most direct response to the stated need for drift detection rather than a compute or logging change.
Why the other options are less suitable
- B — Increase instance size: More compute can improve throughput or latency, but it does not correct a mismatch between the production data distribution and the training data. The accuracy problem is statistical rather than capacity-related.
- C — Reduce log retention: Shorter retention does not detect feature-distribution changes and can reduce evidence available for investigation. The requirement is to monitor data behavior, not to shrink observability data.
- D — Retrain on the same old data: Repeating training on an unchanged historical dataset does not address the new production distribution. Monitoring should first detect and quantify the drift so remediation can use appropriate updated data.
Key takeaway: Detect drift by comparing current production data against a representative baseline and alerting when the distribution moves beyond defined limits.
Preview complete
You have tried all three samples
Review the explanations or try the samples again. This short preview covers only a small part of the exam objectives.
Your sample score resets when this page reloads.
Your practice pack includes
More ways to make each study session count
Focus on a topic, practise under a time limit, or return to the supporting resources before your next attempt.
Domain-based practice
Timed practice sets
Study guide
Flashcards
Requirements
- AWS recommends at least one year using Amazon SageMaker AI, Amazon Bedrock, and other AWS ML-engineering services, plus about one year in a related software, DevOps, data-engineering, or data-science role.
- Candidates should combine question practice with hands-on work involving data pipelines, SageMaker AI, Amazon Bedrock, model training and evaluation, inference, CI/CD, observability, RAG, agents, security, and production troubleshooting.
- You will need an internet-connected device, a current web browser, and an ExamsDigest account to use the online practice product; this does not imply offline availability or a separate mobile application.
Know the exam you’re studying for
These details describe AWS’s official Machine Learning Engineer - Associate (MLA-C02) exam.
What our students say
“The MLA-C02 practice exams helped me find gaps in data preparation, model deployment, and monitoring. The explanations were detailed enough to show why one AWS approach fit the scenario better than the other options.”
Tyler Brennan
✓ Verified Review
“The questions covered a good mix of SageMaker AI, Amazon Bedrock, inference, RAG, model evaluation, and security. Reviewing the questions I missed gave me a much clearer idea of what I still needed to study.”
Sofia Kendrick
✓ Verified Review
“I had already worked through AWS training material before trying ExamsDigest. What I liked most was how the explanations broke down the trade-offs between different deployment and monitoring choices instead of simply showing the correct answer.”
Ethan Rowe
✓ Verified Review
Your questions, answered.
Which Machine Learning Engineer – Associate exam do these questions target?
They target AWS Certified Machine Learning Engineer – Associate MLA-C02, which is currently being delivered as an English-language beta. Beta delivery began September 29, 2026, while general-availability delivery is scheduled for January 14, 2027. If you plan to take the exam after beta, recheck AWS’s exam overview because question count, timing, language availability, or other delivery details can change when the standard exam launches.
Who is MLA-C02 practice intended for?
It is intended for ML engineers, MLOps and LLMOps engineers, data engineers, backend developers, and data scientists who operationalize machine learning and generative AI workloads. AWS recommends at least one year using SageMaker AI, Amazon Bedrock, and related AWS services, plus experience in a related technical role. The guide expects knowledge of traditional ML, foundation models, data pipelines, CI/CD, infrastructure, monitoring, and security rather than beginner-level AI familiarity alone.
How should I combine MLA-C02 practice with hands-on study?
Use practice questions alongside the official guide and hands-on AWS exercises. Useful work includes preparing and validating training data, training SageMaker models, evaluating foundation models, implementing RAG, choosing batch or real-time inference, deploying endpoints, managing model versions, automating CI/CD, observing agents and models, monitoring drift, and configuring IAM and VPC controls. MLA-C02 is implementation-focused, so practical production workflows are important alongside conceptual review.
What should I do when I answer an MLA-C02 question incorrectly?
Use the explanation to identify which engineering decision you missed, then revisit the matching task and skill in the AWS guide. Determine whether the mistake involved data quality, model selection, fine-tuning, RAG, inference type, resource provisioning, pipeline automation, drift, observability, security, or cost. Recreate the scenario in a small AWS exercise where practical. A practice score can highlight weak objectives, but it should not be treated as a guaranteed prediction of an official result.
What question formats appear on the official MLA-C02 exam?
AWS lists multiple-choice and multiple-response questions for MLA-C02. The current beta exam contains 85 questions because AWS includes additional statistical-evaluation items; AWS states that these additional beta items do not affect the score. These three preview examples intentionally use single-answer questions because that is what this widget supports, so they do not reproduce official multiple-response interactions or establish what other formats the paid ExamsDigest product contains.
Does completing this practice earn AWS Certified Machine Learning Engineer – Associate?
No. Completing an ExamsDigest practice product does not award the AWS credential or replace AWS’s official certification process. Practice access, registering for MLA-C02, sitting the AWS beta or later GA exam through an authorized testing option, and receiving the certification are separate activities. AWS says certifications earned through this program remain valid for three years, while MLA-C02 beta and general-availability delivery follow AWS’s published rollout schedule.
30-day conditional refund policy
Eligible requests must be made within 30 calendar days, with no more than 20% of any included simulator completed. The purchase must not already be refunded or under a payment dispute or chargeback. Other provisions apply; mandatory consumer rights remain unaffected.
Useful links for your preparation
