MLA-C01 Question 105
Select 3Your company is implementing a healthcare application that collects sensitive patient data. As an AWS Certified Machine Learning Engineer, you are tasked with ensuring that this data is protected while enabling the data to be used for training a machine learning model. Which of the following techniques should you use to achieve this goal?
- A
Use AWS Glue DataBrew to mask personally identifiable information (PII) in the dataset.
- B
Apply differential privacy techniques to anonymize the dataset.
- C
Encrypt the dataset using AWS Key Management Service (KMS) and train the model directly on encrypted data.
- D
Replace PII fields with synthetic data values using Amazon SageMaker Data Wrangler.
- E
Perform data classification using Amazon Macie to identify and protect sensitive information.
Show answer and explanation
Correct answers: A, B, D
Explanation
To protect sensitive data for machine learning training, you can use data masking tools like AWS Glue DataBrew, anonymization techniques such as differential privacy, or replace sensitive fields with synthetic data. These techniques ensure compliance with data protection regulations while maintaining the dataset's usability. While Amazon Macie helps identify sensitive data, it does not implement masking or anonymization directly. Encrypting the dataset is important for security but does not make the data usable in its encrypted state for model training.
- A. Correct.
AWS Glue DataBrew can be used to mask PII data, making it a suitable option for protecting sensitive information while still enabling its use for machine learning purposes.
- B. Correct.
Differential privacy techniques add noise to the dataset, ensuring anonymity while maintaining its utility for training machine learning models. This is an effective method for protecting sensitive data.
- C. Incorrect.
Encrypting the dataset with AWS KMS is a good security practice, but training a model directly on encrypted data is not feasible due to computational limitations and lack of model interpretability.
- D. Correct.
Replacing PII fields with synthetic data values using Amazon SageMaker Data Wrangler ensures that sensitive information is removed while preserving the dataset's structure, making it useful for training purposes.
- E. Incorrect.
Amazon Macie is a data classification service that identifies sensitive information, but it does not directly anonymize or mask the data. It is a precursor step rather than a standalone solution for this scenario.