MLA-C01 Data Preparation for Machine Learning Practice Test 6 — 15 Questions

Question 1

A company is training a deep learning model on Amazon SageMaker using a dataset stored in Amazon S3. The training job is taking a long time due to I/O bottlenecks. The data is in JSON lines format. Which data preparation step combined with SageMaker's best practices would most effectively reduce training time?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Convert the data to RecordIO-Protobuf format and use SageMaker's Pipe mode for training.. Converting JSON lines data to RecordIO-Protobuf format allows SageMaker's Pipe mode to stream data directly from Amazon S3 to the training algorithm without writing to disk, eliminating I/O bottlenecks. Pipe mode uses a FIFO pipe (named pipe) to feed data sequentially, which significantly reduces training time for deep learning models that iterate over the dataset multiple times.

Answer

Convert the JSON lines files to CSV format and use SageMaker's File mode for training.

Answer

Compress the JSON lines files using gzip and use File mode with local caching.

Answer

Split the data into multiple smaller files and use multiple training instances to parallelize.

Question 2

A data scientist is using Amazon SageMaker Data Wrangler to prepare a dataset. Which TWO features of Data Wrangler can be used to handle imbalanced classification problems? (Choose two.)

Accepted Answer

The Random Oversampling transform to duplicate minority class instances.. Option A is correct because SageMaker Data Wrangler provides a Random Oversampling transform that balances class distribution by duplicating instances of the minority class, directly addressing imbalanced classification. Option B is correct because Data Wrangler also offers a SMOTE (Synthetic Minority Over-sampling Technique) transform, which generates synthetic minority-class samples by interpolating between existing minority instances, a standard approach for imbalanced datasets. Option C is not correct because Drop Duplicates only removes redundant rows and does not rebalance class proportions. Option D is not correct because standardization merely rescales numerical features and has no effect on class imbalance. Option E is not correct because one-hot encoding converts categorical variables into binary vectors and does not address class imbalance.

Answer

The Drop Duplicates transform to remove redundant rows.

Answer

Standardization to scale numerical features.

Answer

One-hot encoding for categorical variables.

Question 3

A company is preparing a large dataset for a SageMaker built-in XGBoost model. The dataset has missing values in both numeric and categorical features, and some categorical features have high cardinality. Which THREE data preparation steps should the company take to optimize model performance? (Choose three.)

Accepted Answer

Split the data into training, validation, and test sets before any imputation.. Option B is correct because splitting into training, validation, and test sets before imputation prevents data leakage: imputation statistics (mean, median, target-encoding mappings) must be learned only from the training set and then applied to validation/test data, otherwise evaluation metrics are optimistically biased. Option C is correct because SageMaker's built-in XGBoost algorithm cannot natively handle missing values in the way some other libraries do, so numeric NaNs should be imputed with a robust statistic such as the median (preferred when outliers exist) or the mean, computed on the training split only. Option D is correct because one-hot encoding is appropriate for low-cardinality categoricals (it avoids imposing false ordinal relationships and keeps dimensionality manageable), while high-cardinality categoricals should use target encoding to avoid an explosion of sparse one-hot columns that would degrade XGBoost performance and memory usage. Option A is not appropriate because outliers can carry legitimate signal for tree-based models like XGBoost, which are relatively robust to them, and blindly removing rows discards data and can bias the model. Option E is incorrect because applying target encoding to all categorical features—including low-cardinality ones—risks overfitting and unnecessary complexity, and target encoding must be done with proper cross-validation/regularization to avoid leakage.

Answer

Remove any rows with outlier values.

Answer

Apply target encoding to all categorical features regardless of cardinality.

Question 4

An e-commerce company uses Amazon SageMaker to train a model that predicts click-through rates. The training data includes a timestamp column 'click_time' and a categorical feature 'device_type' (8 values). They notice that the model's performance degrades over time because the data distribution shifts. They want to ensure the training data represents the most recent behavior. The data is stored in a daily partitioned S3 bucket (e.g., s3://bucket/data/2024-01-01/). The total dataset size is 500 GB. Which approach should they take to prepare the training data while minimizing bias and cost?

Accepted Answer

Select only the data from the last 30 days to train the model.. Selecting only the last 30 days of data directly addresses the data distribution shift by focusing on the most recent user behavior, which is critical for click-through rate prediction. This approach minimizes bias from outdated patterns and reduces training cost by using a smaller, relevant dataset (approximately 500 GB / 365 * 30 ≈ 41 GB). SageMaker training jobs benefit from this reduced volume through faster data loading and lower compute costs.

Answer

Take a random sample of 10% of the rows from the entire dataset.

Answer

Use all historical data and let the model learn the temporal patterns.

Answer

Downsample older data exponentially so that recent data is overrepresented.

Question 5

A data scientist needs to prepare a dataset for a binary classification model. The dataset contains 100,000 records with 50 features, including categorical variables with high cardinality, missing values in 30% of records for a key numeric feature, and a severe class imbalance (5% positive class). The data is stored in an Amazon S3 bucket. Which TWO actions should the data scientist take to improve model performance and ensure robust data preparation? (Choose two.)

Accepted Answer

Use stratified sampling to split the dataset into training and test sets, preserving the class imbalance ratio.. Stratified sampling preserves the class imbalance ratio (5% positive class) in both training and test sets, ensuring the model is evaluated on a representative distribution and avoiding overfitting to the majority class in training. Option E is correct because standardizing numeric features with scikit-learn's StandardScaler in an AWS Glue job ensures that features have zero mean and unit variance, which is critical for distance-based algorithms (e.g., SVM, k-NN) and many neural networks. Option B is incorrect because deleting records with missing values would discard 30% of data, leading to significant information loss and potential bias, and is especially harmful when missingness is non-random. Option C is incorrect because one-hot encoding high-cardinality categorical features creates a very large number of dummy variables, causing the curse of dimensionality and making the model computationally expensive and prone to overfitting. Option D is incorrect because random undersampling of the majority class discards potentially valuable data, often reducing model performance; techniques like SMOTE or class weights are preferred to address class imbalance.

Answer

Delete all records with missing values to ensure data integrity.

Answer

Apply one-hot encoding to all categorical features regardless of cardinality.

Answer

Randomly undersample the majority class to balance the dataset before training.

Question 6

A financial services company is building a fraud detection model using historical transaction data stored in Amazon S3. The data includes features such as transaction amount, merchant category, time of day, and user location. The data scientist observes that the 'merchant_category' column is a text attribute with over 200 unique values. Additionally, the 'transaction_amount' column has a long-tail distribution with extreme outliers. The dataset is 200 GB in size, and the company wants to use Amazon SageMaker for model training. The data scientist needs to engineer features that capture the high-cardinality category and reduce the impact of outliers. What is the MOST efficient and effective approach to prepare this data?

Accepted Answer

Use Amazon EMR with Spark to apply ordinal encoding to merchant_category based on frequency, and log-transform the transaction_amount to reduce skewness.. Ordinal encoding based on frequency handles high-cardinality categorical features efficiently without exploding dimensionality, and log-transform is a standard technique to reduce skewness in long-tail distributions. Using Amazon EMR with Spark provides distributed processing for the 200 GB dataset, making it scalable and cost-effective compared to single-node alternatives.

Answer

Use AWS Glue ETL to apply one-hot encoding to merchant_category and min-max scaling to transaction_amount.

Answer

Use Amazon Athena to bin transaction_amount into 10 equal-width bins and replace merchant_category with its count encoding.

Answer

Use AWS Glue DataBrew to apply a one-hot encoding on merchant_category and a standard scaler on transaction_amount after removing outliers.

Question 7

A healthcare company is developing a predictive model to identify patients at risk of readmission within 30 days after discharge. The dataset contains electronic health record (EHR) data from multiple hospitals, stored as Parquet files in Amazon S3. The data includes patient demographics, diagnoses (ICD-10 codes), medications, lab results, and length of stay. A data scientist notices that the 'lab_result' column has a high number of null values (over 60%) because some tests are not applicable to all patients. Additionally, the 'diagnosis_code' column has over 10,000 unique ICD-10 codes. The company wants to build a model that complies with HIPAA and performs well. The data scientist must prepare the features efficiently using AWS services. Which combination of steps should the data scientist take? (Assume the company can use any AWS service.)

Accepted Answer

Use AWS Glue ETL to impute missing lab results with a value predicted from other features using a model like XGBoost, and apply count encoding to diagnosis codes based on their frequency of occurrence.. It uses AWS Glue ETL to impute missing lab results with a predictive model (XGBoost), which is appropriate for high missingness (>60%) where simple imputation would bias the model, and applies count encoding to the high-cardinality diagnosis codes (10,000+ unique values) to avoid the dimensionality explosion of one-hot encoding while preserving frequency information. This approach balances HIPAA compliance (data stays within AWS) with model performance.

Answer

Replace missing lab results with the overall mean, and use a binary flag for nullness. For diagnosis codes, apply one-hot encoding after grouping codes into 20 categories based on clinical relevance.

Answer

Drop all records where lab_result is null, and use one-hot encoding for diagnosis codes.

Answer

Use Amazon SageMaker Data Wrangler's built-in 'Fill missing' with KNN imputation for lab results, and apply ordinal encoding to diagnosis codes based on the order of ICD-10 chapters.

Question 8

A marketing company is preparing a dataset to train a logistic regression model to predict whether a customer will click on an online ad. The dataset includes 1 million records with features: customer_age (numeric), income (numeric), education_level (ordinal: high school, bachelor, master, PhD), and ad_category (categorical: 50 unique values). The data is stored in a CSV file in Amazon S3. The data scientist plans to use Amazon SageMaker's built-in linear learner algorithm. The data scientist needs to preprocess the data before training. What is the correct sequence of data preparation steps that should be applied to this dataset to ensure optimal model performance?

Accepted Answer

Standardize customer_age and income to have zero mean and unit variance, one-hot encode ad_category, ordinal encode education_level (e.g., map to 1-4), then combine all features into a feature matrix.. It applies appropriate preprocessing for a logistic regression model using SageMaker's linear learner. Standardizing numeric features (zero mean, unit variance) is essential for linear models to ensure convergence and equal feature influence. One-hot encoding the categorical ad_category (50 unique values) avoids imposing ordinal relationships, while ordinal encoding education_level respects its natural order. This combination prepares a feature matrix suitable for the linear learner's optimization.

Answer

Drop any duplicate records, apply min-max scaling to all numeric features, and use target encoding for ad_category based on click rates.

Answer

Apply PCA to all numeric and categorical features after converting categories to numeric indices, then standardize the principal components.

Answer

Apply min-max scaling to customer_age and income, label encode education_level and ad_category, then use recursive feature elimination to reduce dimensionality.

Question 9

A data scientist is preparing a churn dataset in Amazon SageMaker Data Wrangler. The dataset has a numeric column "monthly_spend" with a long right tail, and the downstream XGBoost model is sensitive to feature scale. The scientist wants to reduce skew before training while keeping the transform reproducible in the Data Wrangler flow. Which built-in Data Wrangler transform should be applied to "monthly_spend" to best accomplish this?

Accepted Answer

Log transformation. Monthly spend with a long right tail is a classic case for a variance-stabilizing transform. A log transform compresses the upper tail, making the distribution more symmetric and better behaved for a scale-sensitive algorithm like XGBoost. Because Data Wrangler records the transform as a flow step, the same operation is reproduced when the flow is exported and run as a SageMaker Processing job on new data.

Answer

Standard scaler (z-score normalization)

Answer

Handle outliers

Answer

Encode categorical

Question 10

A machine learning engineer needs to train a model on a dataset stored in Amazon S3 that contains several CSV files. The engineer wants to ensure that the training data is shuffled before being fed to the algorithm. The training job will be run using the SageMaker built-in XGBoost algorithm. Which of the following approaches will MOST efficiently achieve this requirement?

Accepted Answer

Preprocess the data using SageMaker Processing to shuffle the rows and write the shuffled data back to S3 before training.. Shuffling training data is essential to prevent order bias, especially when data is stored in multiple files that may be ordered by time or class. SageMaker Processing jobs can run a custom script to read, shuffle, and rewrite the data to S3, ensuring the algorithm receives randomized input. This method is efficient and works with any algorithm, including built-in ones that lack native shuffle support.

Answer

Enable the SageMaker File mode and rely on the XGBoost algorithm to shuffle the data automatically during training.

Answer

Use SageMaker Augmented Manifest files to specify a shuffle seed, which the algorithm will use to randomize the data.

Answer

Use SageMaker Pipe mode and set the shuffle parameter to true in the algorithm hyperparameters.

Question 11

A data engineer must convert 4 TB of raw JSON event logs stored in Amazon S3 into Parquet for a SageMaker training job. The transformation is a straightforward flattening of nested fields with no joins or user-defined Python logic required. The engineer wants the least operational overhead and the fastest time to production while keeping the output queryable by Amazon Athena. Which approach should the engineer choose?

Accepted Answer

Create an AWS Glue crawler and an AWS Glue ETL job authored with the visual editor, then write the Parquet output back to Amazon S3.. For a simple flattening of nested JSON at terabyte scale with no custom logic, a managed Glue ETL job combined with a crawler delivers the Parquet output and Data Catalog registration with minimal operational effort. The crawler makes the new Parquet data immediately queryable by Athena, and the visual editor avoids hand-written Spark code.

Answer

Run an Amazon Athena CTAS query that reads the JSON and writes the results as Parquet into a new Amazon S3 location.

Answer

Use AWS Lambda to read each JSON object, flatten it, and write Parquet files to Amazon S3.

Answer

Launch an Amazon EMR cluster with Apache Spark, submit a Scala application that flattens the JSON, and terminate the cluster after the run.

Question 12

A data engineer is preparing a dataset for a fraud detection model using Amazon SageMaker Processing. The raw data is stored in Amazon S3 as Parquet files partitioned by date. The processing job needs to read all partitions, apply a transformation, and write the output back to S3 in Parquet format. The engineer wants to minimize the amount of data scanned and ensure the output is partitioned for efficient downstream queries. Which combination of steps should the engineer take?

Accepted Answer

Use the SageMaker Processing Spark container with a script that reads only the relevant partitions, applies the transformation, and writes the output to S3 in Parquet format partitioned by date.. The Spark container in SageMaker Processing supports distributed processing and can read only the necessary partitions, reducing data scanned. Writing output as Parquet partitioned by date allows downstream queries to leverage partition pruning, improving performance. This combination directly addresses the requirements of minimizing scanned data and producing partitioned output for efficient querying.

Answer

Use the SageMaker Processing Spark container with a script that reads the entire S3 prefix, transforms the data, and writes output to a single Parquet file without partitioning.

Answer

Use AWS Glue to crawl the S3 data, create a table in the AWS Glue Data Catalog, and then run an Amazon Athena query to transform the data and write it back to S3 as Parquet, partitioned by date.

Answer

Use the SageMaker Processing scikit-learn container to read the Parquet files, apply the transformation, and write the output as CSV files partitioned by date.

Question 13

A machine learning engineer is preparing a dataset for a model that must be retrained weekly. The raw data arrives in Amazon S3 as Parquet files and is registered in the AWS Glue Data Catalog. They need to create a reproducible, versioned training dataset that can be reused across multiple SageMaker training jobs without re-running the entire extract and transform process each time. Which solution meets these requirements with the LEAST operational overhead?

Accepted Answer

Use a SageMaker Processing job to transform the data and write the output as a SageMaker dataset in S3, then create a training job that references the dataset ARN.. A SageMaker dataset is a versioned, immutable snapshot of data stored in S3, registered with a unique ARN. By using a Processing job to transform the raw data and then registering the output as a dataset, the team creates a reproducible artifact that can be referenced by multiple training jobs without re-running the transform. This meets the requirements with least operational overhead because it uses managed SageMaker capabilities rather than custom versioning.

Answer

Use AWS Glue Studio to transform the data and write the output to a new S3 location, then create a SageMaker training job that reads directly from that location.

Answer

Create a SageMaker Feature Store feature group and ingest the transformed features, then use the offline store for training.

Answer

Use a SageMaker Pipelines workflow that runs a processing job to transform the data, writes the output to S3, and registers a dataset version with SageMaker Lineage and the Model Registry.

Question 14

A data engineer is using AWS Glue to prepare data for a machine learning model. The data is stored in Amazon S3 in Parquet format and partitioned by date. The engineer needs to ensure that the Glue job processes only the most recent partition and that the job can be rerun without duplicating data in the target S3 location. Which TWO actions should the engineer take to meet these requirements? (Choose two.)

Accepted Answer

Configure the job to overwrite the target partition by using dynamic frame write options with partitionBy.. To process only the most recent partition and avoid duplicates on rerun, the engineer should use Glue job bookmarks to track processed data and configure the write operation to overwrite the target partition. Bookmarks ensure that only new partitions are processed, while overwrite mode with partitionBy ensures that reprocessing a partition replaces existing data instead of appending, thus maintaining idempotency.

Answer

Use AWS Glue DataBrew to filter the data by date before writing to the target location.

Answer

Enable auto-scaling for the Glue job to handle the large data volume and reduce processing time.

Answer

Set the job's write mode to 'append' and include a unique identifier for each run to prevent duplicates.

Question 15

A data scientist uses Amazon SageMaker Data Wrangler to prepare features from a 900 GB dataset stored as many small CSV objects in Amazon S3. The visual flow completes but the export step fails with an out-of-memory error, and the resulting SageMaker Processing job log shows the executor running out of memory while reading the source. The team wants to keep using the visual flow with minimal rework. Which change should the data scientist make?

Accepted Answer

Increase the instance count and instance type of the SageMaker Processing job that Data Wrangler runs during export.. The failure is a resource limitation in the Spark-based SageMaker Processing job that Data Wrangler generates at export, not a logic error in the flow. Scaling the processing job's instance type and count gives the executors enough memory and parallelism to handle the many small objects, so the existing visual transformations can be reused unchanged.

Answer

Enable SageMaker Data Wrangler's sampling mode and export only the sampled rows as the training dataset.

Answer

Rewrite the flow as a custom Python script using pandas and run it in a SageMaker Training job.

Answer

Convert the flow to a SageMaker Pipeline and add a Quality Check step before the export.