Free AWS Certified Machine Learning Engineer - Associate (MLA-C01) Practice Questions

Sample MLA-C01 Practice Questions

Browse all 20 free AWS Certified Machine Learning Engineer - Associate practice questions below.

1. A data scientist at a retail company needs to build a product recommendation model. The company has historical purchase data in Amazon S3. The model should be trained using collaborative filtering. Which Amazon SageMaker built-in algorithm is most appropriate for this use case?

   • Amazon SageMaker K-Means for clustering similar users
   • Amazon SageMaker Factorization Machines algorithm, which is optimized for recommendation systems and sparse datasets typical in collaborative filtering
   • Amazon SageMaker Linear Learner for regression on purchase history
   • Amazon SageMaker XGBoost algorithm for gradient boosting on user features

2. An ML engineer is training a deep learning model on Amazon SageMaker. The training job is running on a single ml.p3.2xlarge instance but taking too long. The dataset is 500 GB stored in S3. How can training be accelerated without modifying the model architecture?

   • Convert the dataset to CSV format for faster reading
   • Use SageMaker Pipe Mode to stream data from S3 instead of downloading the full dataset before training
   • Increase the instance type to ml.p3.16xlarge for more CPU cores
   • Enable SageMaker distributed training with data parallelism across multiple instances to split the data across GPUs

3. A company needs to deploy a machine learning model that receives 1000 inference requests per second during peak hours but only 10 requests per second during off-peak. The model takes 200ms per inference. Which SageMaker deployment option best handles this variable load cost-effectively?

   • Use SageMaker Batch Transform for all inference requests
   • Use SageMaker Serverless Inference for automatic scaling to zero
   • Deploy a SageMaker real-time endpoint with application auto scaling configured to scale between 1 and 100 instances based on InvocationsPerInstance metric
   • Deploy on a fixed 100-instance endpoint for peak capacity

4. An ML team wants to track experiments systematically while training multiple model variations. They need to compare metrics, hyperparameters, and artifacts across training runs. Which SageMaker feature provides this experiment tracking capability?

   • SageMaker Model Registry for storing model versions
   • Amazon CloudWatch Metrics for monitoring training jobs
   • SageMaker Debugger for analyzing training issues
   • SageMaker Experiments, which organizes training runs into trials within experiments and tracks metrics, parameters, and artifacts for comparison

5. A financial services company needs to detect fraudulent transactions in near-real-time. They have historical labeled data for model training. The model must provide explainable predictions for regulatory compliance. Which approach meets these requirements?

   • Train an XGBoost model using SageMaker, deploy to a real-time endpoint, and use SageMaker Clarify for feature importance explanations on predictions
   • Use Amazon Fraud Detector without custom models
   • Use a deep neural network for maximum accuracy
   • Deploy using SageMaker Batch Transform for processing

6. A company stores customer features in SageMaker Feature Store. Features include purchase history updated in batch nightly and real-time session features updated during user interactions. How should these different update patterns be handled?

   • Create separate feature groups that must be joined at query time
   • Update all features in real-time only
   • Store real-time features in DynamoDB separately
   • Create a single feature group with both batch and real-time features, using PutRecord API for real-time updates and batch ingestion from S3 for nightly updates

7. A data scientist needs to create features for a time-series forecasting model. The raw data includes sales transactions with timestamps. Which feature engineering techniques are appropriate for time-series data?

   • One-hot encode all timestamp values
   • Create rolling window aggregations (e.g., 30-day moving average)
   • Create lag features capturing values from previous time periods (e.g., sales_7_days_ago)
   • Remove timestamp features entirely as they cause data leakage

8. An ML team uses SageMaker Processing jobs to transform raw data. The transformation involves joining datasets, aggregating features, and applying business logic. Processing takes 4 hours due to data volume. How can processing time be reduced?

   • Increase the instance count in the Processing job configuration for parallel processing
   • Use larger single instances instead of multiple smaller ones
   • Use SageMaker Processing with Spark containers for distributed processing across multiple instances
   • Process only sample data to reduce time

9. A company needs to label medical images for a diagnostic AI system. Due to specialized domain knowledge requirements, only certified radiologists can provide labels. Which SageMaker Ground Truth workforce option is appropriate?

   • Use Amazon Mechanical Turk public workforce for scale
   • Use automated labeling without human review
   • Use AWS-managed third-party vendor workforce
   • Configure a private workforce with the company's certified radiologists as the labeling team

10. A data engineer discovers that a feature in the training data has significant class imbalance—95% of values are in one category. This causes the model to always predict the majority class. How should this imbalance be addressed during data preparation?

    • Apply SMOTE (Synthetic Minority Over-sampling Technique) to generate synthetic examples of the minority class
    • Remove majority class samples to balance the dataset
    • Convert to regression to avoid classification imbalance
    • Use class weights during training to increase the penalty for misclassifying minority class examples

11. A company wants to build a model that predicts customer churn based on tabular data with 50 features. The dataset has 500,000 rows. The team has limited ML expertise. Which SageMaker algorithm is most appropriate?

    • Build a custom deep neural network from scratch
    • Use SageMaker XGBoost, which provides state-of-the-art performance on tabular data with minimal tuning and handles missing values automatically
    • Use SageMaker BlazingText for the prediction
    • Use SageMaker DeepAR for time-series prediction

12. A data scientist trains a classification model and observes high accuracy on training data (98%) but poor accuracy on validation data (72%). Which techniques can address this overfitting?

    • Apply regularization by increasing L1 or L2 penalty to constrain model complexity
    • Use early stopping to halt training when validation performance stops improving
    • Increase model complexity by adding more layers or features
    • Train longer to achieve higher training accuracy

13. A team needs to tune hyperparameters for an XGBoost model. They have budget for 50 training jobs. Which SageMaker hyperparameter tuning strategy explores the parameter space most efficiently?

    • Use Bayesian optimization, which learns from previous results to focus on promising parameter regions and finds good hyperparameters faster than random search
    • Use random search with uniform sampling
    • Use grid search to exhaustively explore all combinations
    • Manually try a few parameter combinations

14. A machine learning engineer needs to train a model on 500GB of image data. A single GPU instance cannot load all data into memory. Which SageMaker capability enables training on this large dataset?

    • Train on a sample and extrapolate results
    • Use a larger single instance with more memory
    • Use SageMaker distributed data parallel training to split the dataset across multiple instances, with each instance processing a subset and gradients synchronized automatically
    • Reduce the dataset to fit in memory

15. A data scientist uses SageMaker Experiments to track training runs. They want to compare multiple model versions across different hyperparameters and datasets. Which Experiments concepts support this?

    • Create separate S3 buckets for each run
    • Create an Experiment to group related trials, where each trial represents a training run with specific hyperparameters and dataset
    • Store all training information in CloudWatch Logs only
    • Log metrics, parameters, and artifacts within each trial using the tracker to capture training details

16. A company needs to deploy a model for real-time predictions with low latency requirements (under 100ms). The model receives sporadic traffic with peaks during business hours. Which SageMaker deployment option best meets these requirements?

    • Use batch transform jobs scheduled hourly
    • Store predictions in a database and serve from there
    • Deploy a real-time SageMaker endpoint with auto-scaling configured to handle traffic variations while maintaining low latency
    • Use SageMaker Processing jobs for each prediction request

17. A data science team needs to score a dataset of 10 million records nightly. Real-time latency is not required. Which SageMaker feature is most cost-effective for this use case?

    • Use SageMaker Batch Transform, which processes the entire dataset without maintaining always-on infrastructure and scales to handle large volumes
    • Use Lambda functions for each prediction
    • Deploy a real-time endpoint and send all records through it
    • Manually process in SageMaker notebooks

18. A company wants to deploy multiple related models that share common preprocessing logic. They want to minimize cold start latency and infrastructure costs. Which deployment pattern achieves this?

    • Combine all models into a single monolithic model
    • Deploy each model to a separate endpoint
    • Use batch transform for all predictions
    • Use a SageMaker multi-model endpoint (MME), which hosts multiple models on shared infrastructure and dynamically loads models as needed

19. A startup has unpredictable traffic patterns—sometimes no requests for hours, then sudden bursts. They want to minimize costs during idle periods while still serving requests quickly. Which deployment option fits best?

    • Deploy a real-time endpoint with minimum instance count of 1
    • Use batch transform triggered by incoming data
    • Manually start and stop endpoints based on schedule
    • Use SageMaker Serverless Inference, which automatically scales from zero during idle periods and spins up capacity for incoming requests

20. A machine learning team wants to gradually roll out a new model version, sending 10% of traffic to the new model while monitoring for issues. Which SageMaker deployment strategy supports this?

    • Use production variants on the same endpoint with traffic weight distribution, assigning 90% to the current model and 10% to the new model
    • Use batch transform to test the new model first
    • Deploy to separate endpoints and use application-level routing
    • Replace the model and roll back if issues occur