Professional Machine Learning Engineer · 13% of the exam

Architecting low-code ML solutions: free practice questions

5 sample questions from our 33-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Google Cloud exam.

1. A Vertex AI AutoML Tabular model trained to predict loan default achieves an AUC-ROC of 0.91 on the evaluation set, but the business team reports that it approves too many defaulting customers (high false negative rate). The dataset has a 5% default rate (severe class imbalance). Which THREE actions would most appropriately address this problem? (Select THREE)

A. Lower the classification threshold used at prediction time to increase sensitivity (recall) for the positive class.✓ Correct
B. Use the class_weight option or apply oversampling of the minority class before retraining.✓ Correct
C. Switch the optimization objective in AutoML Tabular from 'maximize-au-roc' to 'minimize-log-loss' to address imbalance.
D. Evaluate model performance using AUC-PR (precision-recall curve) instead of AUC-ROC, as it is more informative for imbalanced datasets.✓ Correct
E. Retrain the model with 10x more features to increase overall model complexity.
F. Add more majority-class examples to the training dataset to fix the imbalance.

Explanation

Options A, B, and D are correct. (A) Lowering the classification threshold reduces false negatives at the cost of more false positives—directly addressing the business concern of missing defaulters; threshold tuning is the most direct lever after training. (B) Class weighting or oversampling the minority class forces the model to pay more attention to defaulters during training, addressing the root cause of the imbalance problem. (D) AUC-PR is specifically recommended for severely imbalanced datasets because it focuses on the minority class performance, unlike AUC-ROC which can appear artificially high even when the model performs poorly on the minority class. Option C is incorrect: switching optimization objectives from AUC-ROC to log-loss does not specifically target imbalance handling; neither objective directly addresses the class imbalance problem the way resampling or threshold tuning does. Option E is incorrect: adding more features increases complexity and risks overfitting; it does not address class imbalance or the false negative problem. Option F is incorrect: adding more majority-class examples worsens the class imbalance ratio, making the problem worse rather than better.

2. Your team wants to build a model to classify user behavior into 8 distinct engagement tiers using clickstream data stored in BigQuery. You have 500,000 labeled examples and moderate class imbalance. Which approach should you use?

A. BigQuery ML DNN_CLASSIFIER with class_weight configuration to handle imbalance
B. BigQuery ML LINEAR_REG to predict continuous engagement scores, then bucket into tiers
C. Vertex AI AutoML Tabular with the multiclass classification objective and automatic class weighting✓ Correct
D. Pretrained Vision API to analyze user interface interactions

Explanation

Option C is correct because this is a multi-class classification problem (8 tiers), and Vertex AI AutoML Tabular is optimized for tabular multiclass problems at this scale. It automatically handles class imbalance. Option A (DNN_CLASSIFIER) is technically possible but AutoML is the low-code preference for this use case. Option B is incorrect because regression followed by bucketing is suboptimal compared to direct classification. Option D is incorrect because Vision API is for image analysis, not clickstream data.

3. An e-commerce company wants to recommend the next product category for each user. Their data includes user purchase history (500K users, 10M transactions), product metadata (price, category, brand), and user profile features (age, location). They are deciding between three approaches: (1) BigQuery ML matrix factorization, (2) Vertex AI AutoML Tabular predicting the next category, (3) a pretrained recommendation API. Which two approaches are most appropriate for this scenario? (Select TWO)

A. BigQuery ML matrix factorization—handles sparse user-item interaction data without explicit negative samples✓ Correct
B. Vertex AI AutoML Tabular—can use rich user and product features to predict the next category label✓ Correct
C. Vertex AI Recommendations API (pretrained)—optimized for product recommendations with built-in A/B testing
D. BigQuery ML K-means clustering—groups users by purchase patterns, then recommends top products per cluster
E. Vertex AI Vision API—analyzes product images to find visually similar items for recommendation
F. Custom TensorFlow neural collaborative filtering—the only approach that combines implicit feedback with content features

Explanation

Options 1 and 2 are correct. BigQuery ML matrix factorization (option 1) is ideal for collaborative filtering on implicit feedback (purchase history) at scale. Vertex AI AutoML Tabular (option 2) excels when you have rich tabular features (user profile, product metadata) and a discrete target (next category), making it suitable for this multi-feature prediction task. Option 3 (Recommendations API) is a pretrained solution, but it is less flexible for this specific use case where custom features dominate the prediction signal. Option 4 (K-means) is unsupervised clustering and does not directly predict the next category. Option 5 (Vision API) is for image classification, not recommendations. Option 6 (custom TensorFlow) is high-code and not the 'most appropriate' given low-code options exist.

4. A logistics company wants to detect fraudulent shipping claims in real-time as claims arrive. They have 2 years of historical claims data (500,000 records) with 15 features and a 2% fraud rate. Time to prediction must be <100ms. Which solution best balances model accuracy, latency, and operational simplicity?

A. Vertex AI AutoML Tabular with optimized endpoint configuration for low latency✓ Correct
B. BigQuery ML BOOSTED_TREES_CLASSIFIER queried via federated BigQuery ML predictions
C. Custom-trained XGBoost model deployed on Vertex AI with GPU acceleration
D. Pretrained fraud detection API from Google Cloud

Explanation

Option A is correct. Vertex AI AutoML Tabular is optimized for low-latency inference (<100ms) on structured tabular data, and AutoML handles the imbalanced dataset automatically. It provides the best balance of ease (low-code), performance, and latency for this scenario. Option B (BigQuery ML queried via federation) would incur BigQuery query latency, exceeding 100ms. Option C (custom XGBoost) adds operational complexity without guaranteed latency improvement. Option D is incorrect—Google Cloud does not offer a pretrained fraud detection API; this would require custom modeling.

5. A team is deciding between BigQuery ML and Vertex AI AutoML Tabular for predicting customer lifetime value (CLV). Their dataset has 8 million rows, 200 features (mostly text and categorical), and requires monthly retraining. Which two factors most favor AutoML over BigQuery ML in this scenario? (Select TWO)

A. BigQuery ML does not support categorical features natively and requires manual preprocessing
B. Vertex AI AutoML Tabular automatically engineers features from high-cardinality categorical and text columns✓ Correct
C. BigQuery ML BOOSTED_TREES models require explicit hyperparameter tuning, while AutoML automates this✓ Correct
D. AutoML training time is 10x faster than BigQuery ML due to GPU-accelerated training
E. BigQuery ML has a hard limit of 1 million rows for training, while AutoML supports 8 million

Explanation

Options B and C are correct. Option B: AutoML's strength is automatic feature engineering, especially for text and high-cardinality categoricals—a major advantage when you have 200 mixed-type features. Option C: AutoML automates hyperparameter tuning while BigQuery ML BOOSTED_TREES requires manual tuning. Option A is incorrect—BigQuery ML BOOSTED_TREES and LINEAR_REG both support categorical features. Option D is misleading—both scale to 8M rows, and training time depends on data complexity, not a fixed ratio. Option E is false—BigQuery ML has no hard 1M row limit.

28 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free