A. Correct.
Azure Machine Learning provides managed experiment tracking, compute, model registry, pipelines, and deployment features.

B. Incorrect.
Azure DNS hosts DNS zones and does not manage ML experiments.

C. Incorrect.
Azure Files stores file shares but does not provide ML lifecycle management.

D. Incorrect.
Firewall Manager manages network security policy, not data science workflows.

A. Incorrect.
A public IP prefix is networking infrastructure and not an ML training compute target.

B. Incorrect.
Private DNS does not execute notebooks or training jobs.

C. Correct.
Azure ML compute clusters can autoscale for training and reduce nodes when idle to control cost.

D. Incorrect.
Traffic Manager routes DNS traffic and does not run ML workloads.

A. Incorrect.
DNS alias records do not orchestrate ML steps.

B. Correct.
Azure ML pipelines orchestrate repeatable machine learning workflow steps such as preprocessing, training, evaluation, and registration.

C. Incorrect.
Lifecycle rules manage blob tiering and deletion, not model retraining.

D. Incorrect.
WAF policies protect web apps and do not train models.

A. Incorrect.
Blob soft delete protects deleted blobs and is not a modeling technique.

B. Incorrect.
NSG flow logs record network traffic metadata and do not segment users.

C. Incorrect.
DNS forwarding routes name queries and does not build market segments.

D. Correct.
K-means clustering groups observations into similar segments based on feature patterns.

A. Incorrect.
DNS round robin is a traffic distribution pattern, not an ML model.

B. Correct.
Tree-based ensembles can capture non-linear relationships and interactions between features.

C. Incorrect.
A storage queue stores messages and does not learn decision boundaries.

D. Incorrect.
A delete lock protects resources from deletion and is unrelated to classification.

A. Incorrect.
DNS TTL controls name caching and is not a model quality issue.

B. Incorrect.
Gateway transit is a networking peering feature.

C. Correct.
Data drift occurs when production data distributions shift away from training data distributions.

D. Incorrect.
Blob immutability protects data from modification and does not describe distribution shift.

A. Incorrect.
Anonymous access is a security setting, not model monitoring.

B. Incorrect.
VNet peering connects networks and does not retrain models.

C. Incorrect.
DNS failover is unrelated to ML evaluation.

D. Correct.
Monitoring model metrics and triggering retraining when thresholds are breached helps control model degradation.

A. Correct.
Audio features and representative labeled samples help models learn the variation in sounds and phrases.

B. Incorrect.
Removing validation data prevents reliable evaluation.

C. Incorrect.
DNS logs do not contain the audio signal.

D. Incorrect.
Storage tiers affect cost and latency, not recognition quality.

A. Incorrect.
Reducing data usually worsens learning.

B. Correct.
High bias often indicates underfitting, which may be addressed by better features or a more expressive model.

C. Incorrect.
Disabling metrics hides the problem and does not fix it.

D. Incorrect.
DNS is not a machine learning runtime.

A. Incorrect.
A public IP address does not define ML training data.

B. Incorrect.
A CDN rule affects web content delivery.

C. Correct.
Azure ML data assets can version and reuse data definitions across jobs and pipelines.

D. Incorrect.
A Service Bus subscription receives messages and does not version datasets.