MLS-C01 Question 54
Select 3You are tasked with building a machine learning model to predict customer churn for an e-commerce company. The raw dataset contains categorical features (e.g., customer segment, region) and numerical features (e.g., age, total spend). The numerical features contain missing values, and the categorical features include high-cardinality variables. You need to prepare the data for training while ensuring that the transformations are optimal for machine learning workflows on AWS. What steps should you take to transform the data?
- A
Use Amazon SageMaker Data Wrangler to impute missing numerical values and encode categorical features with high cardinality using target encoding.
- B
Normalize numerical features to a standard scale using Min-Max scaling and one-hot encode all categorical features using Amazon SageMaker Processing jobs.
- C
Impute missing numerical values with the mean and apply feature hashing to high-cardinality categorical features using AWS Glue.
- D
Drop rows with missing numerical values and use Amazon SageMaker Feature Store to one-hot encode categorical features.
- E
Use Amazon SageMaker Feature Store to store transformed data, enabling feature reuse across multiple ML models.
Show answer and explanation
Correct answers: A, C, E
Explanation
Transforming data for machine learning on AWS requires identifying the optimal tools and techniques for the dataset's characteristics. High-cardinality categorical features should be encoded using efficient methods like target encoding or feature hashing, while missing numerical values can be imputed with statistical measures like the mean. Amazon SageMaker Data Wrangler, AWS Glue, and SageMaker Feature Store are essential services for implementing scalable and reusable data transformation workflows, making options 1, 3, and 5 correct.
- A. Correct.
This option is correct because Amazon SageMaker Data Wrangler provides a no-code interface to handle common data transformation tasks like imputing missing values and target encoding, which is suitable for high-cardinality categorical features.
- B. Incorrect.
This option is incorrect because one-hot encoding all categorical features is not efficient for high-cardinality variables. Additionally, Min-Max scaling may not be ideal for all ML models, depending on their sensitivity to feature scaling.
- C. Correct.
This option is correct because imputing missing values with the mean is a standard approach, and feature hashing is an efficient encoding method for high-cardinality categorical features. AWS Glue supports data preparation and transformation at scale.
- D. Incorrect.
This option is incorrect because dropping rows with missing numerical values can lead to data loss, especially if the dataset is small. Also, one-hot encoding high-cardinality categorical features is not optimal.
- E. Correct.
This option is correct because Amazon SageMaker Feature Store enables centralized storage of transformed features, making it easier to reuse them across different ML models and pipelines.