100% Latest & Updated Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test 17
Question 160
During risk sign-off, the team must assess this approved design step: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key. This relates to a situation where a regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. Which limitation is most important when deciding whether this change is suitable?
- Stratification cannot prevent leakage from related records belonging to the same customer.
- High raw agreement can hide the absence of meaningful minority class examples.
- The processing role also needs KMS permissions for decrypting authorized objects.
- Block Public Access does not replace least-privilege private access review.
Explanation
Correct Answer
C:This is a real operational limit for the described design: The processing role also needs KMS permissions for decrypting authorized objects. The associated control is: Use S3 server-side encryption with an appropriately restricted customer-managed KMS key. Its intended result is: Stored training objects are encrypted with an auditable key access boundary. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: Only 2 percent of historical medical claims are fraudulent, and random splitting sometimes leaves too few fraud examples in evaluation data. The relevant boundary for this case is: The processing role also needs KMS permissions for decrypting authorized objects. These are different constraints.
B:That concern belongs to a different design: A text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. The relevant boundary for this case is: The processing role also needs KMS permissions for decrypting authorized objects. These are different constraints.
D:That concern belongs to a different design: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The relevant boundary for this case is: The processing role also needs KMS permissions for decrypting authorized objects. These are different constraints.
Question 161
A release gate must account for this limitation while using: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals. This relates to a situation where model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. Which limitation is most important when deciding whether this change is suitable?
- Removing every repeated event can destroy genuine time-series information, so the deduplication key matters.
- A catalog entry alone is insufficient if actual input object versions were not retained.
- Stratification cannot prevent leakage from related records belonging to the same customer.
- Block Public Access does not replace least-privilege private access review.
Explanation
Correct Answer
D:This is a real operational limit for the described design: Block Public Access does not replace least-privilege private access review. The associated control is: Enable S3 Block Public Access and apply a restricted bucket policy for approved principals. Its intended result is: Public ACL or policy misconfigurations cannot expose the protected dataset through blocked public access paths. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: The same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. The relevant boundary for this case is: Block Public Access does not replace least-privilege private access review. These are different constraints.
B:That concern belongs to a different design: An auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. The relevant boundary for this case is: Block Public Access does not replace least-privilege private access review. These are different constraints.
C:That concern belongs to a different design: Only 2 percent of historical medical claims are fraudulent, and random splitting sometimes leaves too few fraud examples in evaluation data. The relevant boundary for this case is: Block Public Access does not replace least-privilege private access review. These are different constraints.
Question 162
The technical review is assessing an important boundary of this proposal: Enforce schema and type validation at ingestion and quarantine incompatible records. This relates to a situation where a feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. Which limitation is most important when deciding whether this change is suitable?
- A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role.
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
- Block Public Access does not replace least-privilege private access review.
- Accepting every string as convertible can silently change feature meaning.
Explanation
Correct Answer
D:This is a real operational limit for the described design: Accepting every string as convertible can silently change feature meaning. The associated control is: Enforce schema and type validation at ingestion and quarantine incompatible records. Its intended result is: Unexpected feature types are rejected before they reach the serving pipeline. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: An international organization must train a model using regional customer data without copying identifiable records into another AWS Region. The relevant boundary for this case is: Accepting every string as convertible can silently change feature meaning. These are different constraints.
B:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: Accepting every string as convertible can silently change feature meaning. These are different constraints.
C:That concern belongs to a different design: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The relevant boundary for this case is: Accepting every string as convertible can silently change feature meaning. These are different constraints.
Question 163
During risk sign-off, the team must assess this approved design step: Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage. This relates to a situation where an auditor must trace a prediction feature back to its original S3 object, Glue job run, and data processing version after a model incident. Which limitation is most important when deciding whether this change is suitable?
- Stratification cannot prevent leakage from related records belonging to the same customer.
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
- Block Public Access does not replace least-privilege private access review.
- A catalog entry alone is insufficient if actual input object versions were not retained.
Explanation
Correct Answer
D:This is a real operational limit for the described design: A catalog entry alone is insufficient if actual input object versions were not retained. The associated control is: Record immutable input references, job identifiers, schemas, and transformation versions in the data catalog lineage. Its intended result is: Incident reviewers can reconstruct which source data and processing steps produced a feature set. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: Only 2 percent of historical medical claims are fraudulent, and random splitting sometimes leaves too few fraud examples in evaluation data. The relevant boundary for this case is: A catalog entry alone is insufficient if actual input object versions were not retained. These are different constraints.
B:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: A catalog entry alone is insufficient if actual input object versions were not retained. These are different constraints.
C:That concern belongs to a different design: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The relevant boundary for this case is: A catalog entry alone is insufficient if actual input object versions were not retained. These are different constraints.
Question 164
A release gate must account for this limitation while using: Refine annotation instructions, adjudicate ambiguous examples, and relabel affected training data. This relates to a situation where a text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. Which limitation is most important when deciding whether this change is suitable?
- High raw agreement can hide the absence of meaningful minority class examples.
- A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role.
- The processing role also needs KMS permissions for decrypting authorized objects.
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
Explanation
Correct Answer
A:This is a real operational limit for the described design: High raw agreement can hide the absence of meaningful minority class examples. The associated control is: Refine annotation instructions, adjudicate ambiguous examples, and relabel affected training data. Its intended result is: Training targets become more internally consistent for model fitting. This limitation must still be managed.
Incorrect Answers
B:That concern belongs to a different design: An international organization must train a model using regional customer data without copying identifiable records into another AWS Region. The relevant boundary for this case is: High raw agreement can hide the absence of meaningful minority class examples. These are different constraints.
C:That concern belongs to a different design: A regulated ML team stores customer data in S3 and wants control over encryption-key access by execution role and audited key usage. The relevant boundary for this case is: High raw agreement can hide the absence of meaningful minority class examples. These are different constraints.
D:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: High raw agreement can hide the absence of meaningful minority class examples. These are different constraints.
Question 165
The technical review is assessing an important boundary of this proposal: Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified. This relates to a situation where a training job repeatedly reads a shared block of feature matrices and needs much faster parallel data access than the current single-volume setup. Which limitation is most important when deciding whether this change is suitable?
- A passing schema check cannot guarantee every record's business meaning is correct.
- Storage acceleration does not correct poor model code or a CPU bottleneck.
- Removing every repeated event can destroy genuine time-series information, so the deduplication key matters.
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
Explanation
Correct Answer
B:This is a real operational limit for the described design: Storage acceleration does not correct poor model code or a CPU bottleneck. The associated control is: Choose a compatible shared high-performance training filesystem such as FSx for Lustre when justified. Its intended result is: Training workers can access the prepared matrices concurrently at adequate throughput. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: A data lake receives a daily product feed that must have nonnull customer keys and positive amounts before it enters feature processing. The relevant boundary for this case is: Storage acceleration does not correct poor model code or a CPU bottleneck. These are different constraints.
C:That concern belongs to a different design: The same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. The relevant boundary for this case is: Storage acceleration does not correct poor model code or a CPU bottleneck. These are different constraints.
D:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: Storage acceleration does not correct poor model code or a CPU bottleneck. These are different constraints.
Question 166
During risk sign-off, the team must assess this approved design step: Deduplicate appropriate rows and group records by entity before assigning train/test partitions. This relates to a situation where the same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. Which limitation is most important when deciding whether this change is suitable?
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
- Block Public Access does not replace least-privilege private access review.
- Removing every repeated event can destroy genuine time-series information, so the deduplication key matters.
- A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role.
Explanation
Correct Answer
C:This is a real operational limit for the described design: Removing every repeated event can destroy genuine time-series information, so the deduplication key matters. The associated control is: Deduplicate appropriate rows and group records by entity before assigning train/test partitions. Its intended result is: Evaluation reduces accidental leakage of the same customer's behavior across partitions. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: Removing every repeated event can destroy genuine time-series information, so the deduplication key matters. These are different constraints.
B:That concern belongs to a different design: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The relevant boundary for this case is: Removing every repeated event can destroy genuine time-series information, so the deduplication key matters. These are different constraints.
D:That concern belongs to a different design: An international organization must train a model using regional customer data without copying identifiable records into another AWS Region. The relevant boundary for this case is: Removing every repeated event can destroy genuine time-series information, so the deduplication key matters. These are different constraints.
Question 167
A release gate must account for this limitation while using: Apply suitable class weights or resampling on training data and evaluate minority recall or F1. This relates to a situation where a defective-parts classifier predicts the majority good class for nearly every example because defective units are rare and accuracy appears high. Which limitation is most important when deciding whether this change is suitable?
- Resampling evaluation or test data changes the prevalence represented in reported metrics.
- A numerical bias score is not proof of legal fairness; Clarify is available only to existing customers after July 30, 2026.
- A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role.
- Block Public Access does not replace least-privilege private access review.
Explanation
Correct Answer
A:This is a real operational limit for the described design: Resampling evaluation or test data changes the prevalence represented in reported metrics. The associated control is: Apply suitable class weights or resampling on training data and evaluate minority recall or F1. Its intended result is: Training places meaningful importance on rare defects rather than only majority accuracy. This limitation must still be managed.
Incorrect Answers
B:That concern belongs to a different design: A lender already using SageMaker Clarify has much fewer approved loans from one demographic subgroup and needs a measured dataset bias check before modeling. The relevant boundary for this case is: Resampling evaluation or test data changes the prevalence represented in reported metrics. These are different constraints.
C:That concern belongs to a different design: An international organization must train a model using regional customer data without copying identifiable records into another AWS Region. The relevant boundary for this case is: Resampling evaluation or test data changes the prevalence represented in reported metrics. These are different constraints.
D:That concern belongs to a different design: Model input CSV files are written into an S3 bucket that mistakenly has a permissive ACL, and compliance forbids anonymous access regardless of the object setting. The relevant boundary for this case is: Resampling evaluation or test data changes the prevalence represented in reported metrics. These are different constraints.
Question 168
The technical review is assessing an important boundary of this proposal: Keep raw customer data and processing resources in the permitted AWS Region with restrictive transfer controls. This relates to a situation where an international organization must train a model using regional customer data without copying identifiable records into another AWS Region. Which limitation is most important when deciding whether this change is suitable?
- High raw agreement can hide the absence of meaningful minority class examples.
- A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role.
- Removing every repeated event can destroy genuine time-series information, so the deduplication key matters.
- Accepting every string as convertible can silently change feature meaning.
Explanation
Correct Answer
B:This is a real operational limit for the described design: A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role. The associated control is: Keep raw customer data and processing resources in the permitted AWS Region with restrictive transfer controls. Its intended result is: The workflow respects the specified regional data-processing boundary. This limitation must still be managed.
Incorrect Answers
A:That concern belongs to a different design: A text moderation classifier has confusing class definitions and annotators disagree about which of two labels applies to borderline tickets. The relevant boundary for this case is: A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role. These are different constraints.
C:That concern belongs to a different design: The same customer appears in nearly identical records across training and evaluation splits, yielding unrealistic customer churn accuracy. The relevant boundary for this case is: A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role. These are different constraints.
D:That concern belongs to a different design: A feature producer changes a numeric transaction count to a string, causing model inference preprocessing to fail despite successful S3 ingestion. The relevant boundary for this case is: A local bucket does not prevent unauthorized cross-Region transfers by a broadly privileged role. These are different constraints.
Question 169
A single ML platform team is handling two distinct changes for the same system. Workstream A: A financial insurer predicts whether a structured claim is fraudulent using numerical and categorical features, including nonlinear interactions and missing values. Workstream B: A manufacturer has machine vibration signals but only a handful of labeled failures and wants to flag unusual sensor readings. Select exactly TWO measures, one that directly resolves each workstream.
- Call Amazon Comprehend sentiment analysis for supported text and language requirements.
- Use a supported unsupervised anomaly detection approach such as Random Cut Forest.
- Use Amazon Translate for supported language pairs with required terminology controls.
- Prefer an interpretable linear classifier with documented preprocessing and feature contribution analysis.
- Select a tabular gradient-boosted tree model such as SageMaker XGBoost for the structured classification task.
Explanation
Correct Answers
B:This action fits the situation: A manufacturer has machine vibration signals but only a handful of labeled failures and wants to flag unusual sensor readings. It implements the needed technical control: Use a supported unsupervised anomaly detection approach such as Random Cut Forest. The intended outcome is that unusual observations can be ranked for investigation without extensive failure labels. The important limitation is that an anomaly score does not by itself prove a mechanical fault or determine root cause.
E:This action fits the situation: A financial insurer predicts whether a structured claim is fraudulent using numerical and categorical features, including nonlinear interactions and missing values. It implements the needed technical control: Select a tabular gradient-boosted tree model such as SageMaker XGBoost for the structured classification task. The intended outcome is that the classifier can learn nonlinear interactions without a manually engineered deep network. The important limitation is that tree models still need careful leakage controls and can overfit sparse customer subgroups.
Incorrect Answers
A:That alternative addresses a different task: A support organization wants general sentiment scores from English customer comments and lacks custom labels or ML engineering staff. The selected workloads instead require Gradient boosted claims model, Rare event anomaly discovery.
C:That alternative addresses a different task: A global service must translate short customer support responses from Spanish to English without creating parallel-text training data. The selected workloads instead require Gradient boosted claims model, Rare event anomaly discovery.
D:That alternative addresses a different task: A compliance-sensitive classifier needs understandable positive and negative influences of a few engineered features, and a deep opaque model is unnecessary. The selected workloads instead require Gradient boosted claims model, Rare event anomaly discovery.
Question 170
A single ML platform team is handling two distinct changes for the same system. Workstream A: A transport firm predicts delivery cost as a continuous numeric amount and requires a simple coefficient-based explanation for business reviewers. Workstream B: A content team needs general-purpose object and scene labels for uploaded catalog pictures but does not have enough examples to train a custom computer-vision model. Select exactly TWO measures, one that directly resolves each workstream.
- Start with a regularized linear regression model and validate predictive assumptions.
- Use Amazon Personalize with an appropriate dataset group and recommender configuration.
- Use a managed image-labeling API such as Rekognition for supported generic categories.
- Prefer an interpretable linear classifier with documented preprocessing and feature contribution analysis.
- Use logistic regression with calibrated probabilities as an interpretable binary-classification baseline.
Explanation
Correct Answers
A:This action fits the situation: A transport firm predicts delivery cost as a continuous numeric amount and requires a simple coefficient-based explanation for business reviewers. It implements the needed technical control: Start with a regularized linear regression model and validate predictive assumptions. The intended outcome is that predictions and model coefficients support an auditable numeric cost baseline. The important limitation is that strong nonlinearities or correlated predictors can make a linear specification inaccurate or unstable.
C:This action fits the situation: A content team needs general-purpose object and scene labels for uploaded catalog pictures but does not have enough examples to train a custom computer-vision model. It implements the needed technical control: Use a managed image-labeling API such as Rekognition for supported generic categories. The intended outcome is that the application obtains image tags without maintaining custom training infrastructure. The important limitation is that generic pretrained labels may not recognize specialized products or fulfill strict domain accuracy.
Incorrect Answers
B:That alternative addresses a different task: A streaming-media business wants recommendations based on user-item interaction histories without designing its own collaborative-filtering architecture. The selected workloads instead require Interpretable linear pricing prediction, Managed image classification.
D:That alternative addresses a different task: A compliance-sensitive classifier needs understandable positive and negative influences of a few engineered features, and a deep opaque model is unnecessary. The selected workloads instead require Interpretable linear pricing prediction, Managed image classification.
E:That alternative addresses a different task: A lender needs a probability that an applicant will default, not a continuous estimated loan amount, and requires threshold-based risk decisions. The selected workloads instead require Interpretable linear pricing prediction, Managed image classification.
AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Tests
