Microsoft Certified: Azure Data Scientist Associate Intermediate Practice Exam: Medium Difficulty 2025

You are designing a machine learning solution to predict customer churn for a retail company. The dataset contains 50,000 records with 30 features, including customer demographics, purchase history, and engagement metrics. However, only 5% of customers have churned. Which combination of techniques should you implement to address the class imbalance and ensure reliable model performance?

You are training a regression model in Azure Machine Learning to predict housing prices. During exploratory data analysis, you discover that several numeric features have different scales: square footage (500-5000), number of bedrooms (1-6), and price per square foot (50-500). The dataset also contains missing values in the 'year_renovated' column. What is the correct sequence of preprocessing steps to implement in your Azure ML pipeline?

Your team has trained multiple models using Azure Machine Learning AutoML for a classification task. You need to select the best model for deployment that balances performance and inference latency requirements. The production environment requires predictions to be returned within 100ms. Which approach should you take to evaluate and select the appropriate model?

You have deployed a machine learning model as a web service on Azure Kubernetes Service (AKS) using Azure Machine Learning. After deployment, you notice that the service occasionally returns HTTP 503 errors during peak hours. Application Insights shows that CPU utilization reaches 95% during these periods. What is the most effective solution to resolve this issue?

You are implementing hyperparameter tuning for a deep learning model in Azure Machine Learning. The model has five hyperparameters to optimize, and each training run takes approximately 45 minutes. You have a limited compute budget and need to find optimal hyperparameters within 24 hours. Which sampling method and early termination policy combination would be most appropriate?

Your organization requires that all machine learning models must be explainable to comply with regulatory requirements. You have trained a gradient boosting model using Azure Machine Learning for a loan approval system. Which combination of techniques should you use to provide both global and local model interpretability?

You are designing a data preparation pipeline in Azure Machine Learning for a time-series forecasting project. The pipeline needs to process incoming data from Azure SQL Database, perform feature engineering, and handle incremental data updates daily. Which Azure ML components should you use to build an efficient and maintainable solution?

You have deployed a real-time inference endpoint for a classification model in Azure Machine Learning. After two weeks in production, you notice that the model's precision has dropped from 0.85 to 0.72. You need to implement a monitoring solution to detect and alert on model performance degradation. What should you implement?

You are training a computer vision model using Azure Machine Learning to classify product defects. The training dataset contains 100,000 images stored in Azure Blob Storage. Training is taking longer than expected. Which approach would most effectively improve training performance without significantly impacting model accuracy?

You need to deploy a machine learning model that will be consumed by a mobile application. The application requires offline capability, meaning predictions must be made on-device without internet connectivity. The model is a scikit-learn random forest classifier. Which deployment approach should you recommend?