Machine Learning Engineer Practice Exam: Test Your Knowledge 2025

A retail company wants to predict customer churn based on historical data. They have labeled data showing which customers left in the past 2 years. The business wants to identify at-risk customers monthly to offer retention incentives. Which type of ML problem is this?

Your team is building a recommendation system for an e-commerce platform. The business team asks for a model that can explain why specific products are recommended to users for regulatory compliance. Which approach should you prioritize?

A healthcare provider wants to detect diseases from medical images. They have 500 labeled images but need a production-ready model. The domain experts indicate that similar research models exist. What is the most efficient approach to frame this problem?

Your organization needs to deploy a real-time fraud detection model that processes credit card transactions. The model must return predictions within 100ms and handle 10,000 requests per second. Which deployment architecture is most appropriate?

You are designing an ML solution for a global application that must comply with data residency requirements in multiple regions. Training data cannot leave specific geographic boundaries. How should you architect the solution?

A financial services company needs to serve ML predictions while ensuring model artifacts and data are encrypted both at rest and in transit. Which Google Cloud services combination provides this security?

Your team needs to perform hyperparameter tuning for a complex deep learning model. Training a single model takes 4 hours. You need to evaluate 100 different hyperparameter combinations. What is the most cost-effective approach?

You need to prepare a dataset containing user activity logs stored in Cloud Storage as JSON files. The data needs to be transformed, validated, and split into training and evaluation sets at scale. Which tool is most appropriate?

Your ML pipeline ingests streaming data from IoT devices via Pub/Sub. You need to perform feature engineering with a 10-minute sliding window aggregation before feeding data to your model. What architecture should you use?

You have a dataset with significant class imbalance: 95% of examples are negative, 5% are positive. The business considers false negatives 10x more costly than false positives. How should you prepare the data?

Your team is building features from multiple data sources: BigQuery tables, Cloud SQL database, and real-time streaming data. You need to serve consistent features for both training and online prediction. What solution should you implement?

You are training a custom TensorFlow model on Vertex AI with a large dataset. Training is taking too long. Which optimization strategies should you implement first?

You need to develop a text classification model. Your team has limited ML expertise and needs a solution deployed quickly. Which approach is most appropriate?

During model evaluation, your classification model shows high precision (0.95) but low recall (0.45) on the validation set. The business needs to identify as many positive cases as possible. What should you do?

You are training a regression model to predict housing prices. The model performs well on training data (R² = 0.92) but poorly on validation data (R² = 0.58). What is the most likely issue and solution?

You need to implement a recommendation model that learns from both user-item interactions and user/item features. The system must handle millions of users and items efficiently. Which architecture is most appropriate?

Your organization needs to automate the retraining of an ML model whenever new labeled data is added to BigQuery. The pipeline should handle data validation, model training, evaluation, and deployment if quality thresholds are met. What should you use?

You have multiple ML models in production with different training schedules, data dependencies, and deployment conditions. You need a solution that provides versioning, lineage tracking, and reproducibility. What architecture should you implement?

Your production model's prediction accuracy has decreased from 0.88 to 0.76 over the past month. What is the most likely cause and immediate action?

You need to implement comprehensive monitoring for a production ML model serving predictions via Vertex AI. What metrics and monitoring approach should you implement?