IBM A1000-080: Assessment: Data Science and AI Intermediate Practice Exam: Medium Difficulty 2025

A data scientist is analyzing customer purchase data and notices that 15% of the records have missing values in the 'income' column. The income data appears to be missing not at random (MNAR), as higher-income customers tend to skip this question. Which imputation strategy would be most appropriate for handling this missing data?

A retail company wants to predict customer churn using historical transaction data. The dataset is highly imbalanced with only 5% of customers churning. After training a logistic regression model, it achieves 95% accuracy but fails to identify most churners. What combination of techniques would best address this issue?

During model development, a data scientist observes that the training error is 2% while the validation error is 18%. The model performs well on training data but poorly on new data. What is the most effective approach to address this problem?

A healthcare organization needs to build a model to predict patient readmission risk. The model must provide explanations for its predictions to comply with regulations and gain physician trust. Which modeling approach would best balance predictive performance with interpretability?

A data science team is performing feature engineering on a dataset with multiple categorical variables including 'city' (2000 unique values) and 'product_category' (50 unique values). They need to prepare these features for a random forest model. What encoding strategy would be most appropriate?

An AI team is building a convolutional neural network (CNN) for image classification. During training, they notice the validation loss starts increasing after epoch 15 while training loss continues to decrease. The model has 5 convolutional layers and 3 fully connected layers. What combination of techniques would most effectively address this issue?

A company is deploying a natural language processing model for sentiment analysis in production. The model was trained on customer reviews from 2020-2021 but will process reviews from 2024 onwards. What approach should be implemented to maintain model performance over time?

A data scientist is working with a dataset containing both numerical features (age, income, credit score) and text features (customer feedback). They need to build a unified model that leverages both data types. Which architectural approach would be most effective?

A machine learning project requires processing a dataset with 500,000 records and 1,000 features. The team needs to reduce dimensionality while preserving the most important information for a classification task. What approach would be most appropriate?

An organization is implementing an AI governance framework for their data science projects. They need to ensure models are fair, transparent, and compliant with regulations. Which combination of practices should be prioritized in their MLOps pipeline?