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

Sample MLA-C01 Practice Questions

Browse all 10 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