A government agency has a dataset of citizen service requests that includes free-text descriptions, categorical fields (service type, district), and numeric fields (response time). The agency needs to build a model to predict service request priority. The data contains PII in the free-text descriptions (e.g., names, addresses). The team must remove PII while preserving the text's usefulness for classification. What is the most appropriate approach?

Use a Cloud DLP inspection job to redact PII from the text, then use the cleaned text for feature extraction

Remove the free-text column entirely and train only on categorical and numeric fields

Hash the entire free-text description to anonymize it

Manually review and edit all text descriptions to remove PII

A human resources company is building a model to predict employee attrition. The dataset includes employee names, email addresses, and performance ratings. The privacy team requires that employee names and email addresses be removed from the dataset before it is used for training. What is the most appropriate preprocessing step?

Drop the name and email columns and keep the remaining features

Hash the name and email columns and keep them in the dataset

Replace names and emails with random IDs and keep them in the dataset

Encrypt the name and email columns and keep them in the dataset

A multinational company is building a global customer analytics model. The data from EU customers must comply with GDPR, which requires that personal data not be transferred outside the EU. The ML team needs to preprocess data in a way that allows a global model to be trained without moving EU personal data. What is the most appropriate approach?

Train separate models for EU and non-EU data and never combine them

Copy all EU data to a US region and train the global model there

Anonymize the EU data by removing names and email addresses, then copy it globally

Preprocess EU data in an EU region, extract non-personal aggregate features, and share only those features globally

A large online retailer has a feature 'user_session_duration' that is computed by a batch pipeline every hour and also by a streaming pipeline in near real-time. The ML team notices that the online and offline models produce inconsistent predictions because the feature values differ between the two pipelines. They want to ensure consistency without sacrificing the near real-time requirement. What should they do?

Use Gemini Enterprise Agent Platform Feature Store with a single streaming pipeline that serves both online and offline

Use the batch pipeline for both online and offline, accepting the hourly latency

Switch the offline training to use the batch pipeline and the online serving to use the streaming pipeline

Keep both pipelines but add a reconciliation job that overwrites the offline values with online values

A manufacturing company has sensor data from IoT devices stored in BigQuery. The data includes timestamps, sensor readings, and device IDs. The ML team wants to build a predictive maintenance model. They notice that some sensors have missing values and others have outliers. What is the most appropriate way to organize and explore this time-series data before feature engineering?

Create a feature store entry for each sensor reading without any exploration

Use BigQuery to compute summary statistics and visualize time-series patterns with Looker Studio

Export the data to CSV and load it into Excel for manual inspection

Immediately train a model and use feature importance to identify problematic sensors

Google CloudProfessional Machine Learning Engineer

Domain 2Objective 1

Exploring and Preprocessing Data for ML PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 4)

Part of the Collaborating within and across teams to manage data and models domain, which accounts for ~16% of the PROFESSIONAL-MACHINE-LEARNING-ENGINEER exam.

24questions here

5free pages

4concepts

~16%of the exam

Questions 16–20

16expert · hard
A government agency has a dataset of citizen service requests that includes free-text descriptions, categorical fields (service type, district), and numeric fields (response time). The agency needs to build a model to predict service request priority. The data contains PII in the free-text descriptions (e.g., names, addresses). The team must remove PII while preserving the text's usefulness for classification. What is the most appropriate approach?
Select an answer first
17application · medium
A human resources company is building a model to predict employee attrition. The dataset includes employee names, email addresses, and performance ratings. The privacy team requires that employee names and email addresses be removed from the dataset before it is used for training. What is the most appropriate preprocessing step?
Select an answer first
18expert · hard
A multinational company is building a global customer analytics model. The data from EU customers must comply with GDPR, which requires that personal data not be transferred outside the EU. The ML team needs to preprocess data in a way that allows a global model to be trained without moving EU personal data. What is the most appropriate approach?
Select an answer first
19expert · hard
A large online retailer has a feature 'user_session_duration' that is computed by a batch pipeline every hour and also by a streaming pipeline in near real-time. The ML team notices that the online and offline models produce inconsistent predictions because the feature values differ between the two pipelines. They want to ensure consistency without sacrificing the near real-time requirement. What should they do?
Select an answer first
20application · medium
A manufacturing company has sensor data from IoT devices stored in BigQuery. The data includes timestamps, sensor readings, and device IDs. The ML team wants to build a predictive maintenance model. They notice that some sensors have missing values and others have outliers. What is the most appropriate way to organize and explore this time-series data before feature engineering?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to PROFESSIONAL-MACHINE-LEARNING-ENGINEER

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Google Cloud. “PROFESSIONAL-MACHINE-LEARNING-ENGINEER” is a trademark of its owner, used for identification only.

Exploring and Preprocessing Data for ML PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 4)

Questions 16–20

Keep studying — Collaborating within and across teams to manage data and models