IBM A1000-041 - Assessment: Data Science Foundations - Level 1 Advanced Practice Exam: Hard Questions 2025

A data science team is implementing the CRISP-DM methodology for a customer churn prediction project. During the modeling phase, they discover that their initial business understanding was flawed - the cost of false negatives (missing churners) is 10x higher than false positives (incorrectly predicting churn). They've already completed data preparation with balanced sampling. What is the MOST appropriate next step according to CRISP-DM principles?

During the data understanding phase of a healthcare analytics project, you discover that 35% of patient records have missing values for a critical biomarker, but the missingness is not random - it's systematically missing for patients from lower-income zip codes who lack certain insurance coverage. The biomarker is strongly predictive of the outcome. Which approach BEST addresses both the analytical and ethical implications?

You're visualizing the relationship between three continuous variables (sales, advertising spend, and customer satisfaction) across 8 different product categories over 24 months. Stakeholders need to understand temporal trends, cross-category comparisons, and multivariate relationships simultaneously. Which visualization strategy would be MOST effective for this complex analytical requirement?

You're analyzing a dataset with extreme right skewness in the target variable (income, ranging from $20K to $50M with 99% of values below $200K). You've applied log transformation for modeling, achieved good metrics on test data (R²=0.82), but when you inverse-transform predictions back to the original scale, stakeholders report the predictions are systematically underestimating actual values for high earners. What is the MOST likely cause and appropriate solution?

You need to create an automated visualization pipeline that generates reports for 150+ regional managers, each showing their region's performance. The data includes sensitive financial metrics with different access levels. Some regions have sparse data (<10 records) while others have thousands. Which design approach BEST balances automation, statistical validity, and data governance?

You're working with a large pandas DataFrame (50M rows, 200 columns) and need to perform a complex operation: group by 3 categorical columns, calculate rolling 7-day averages within each group, then merge back calculated z-scores for each group. Memory errors are occurring. Which optimization strategy provides the BEST combination of performance and memory efficiency?

You're building a data processing pipeline that reads JSON from an API, performs transformations, and outputs to multiple formats. The API occasionally returns malformed JSON, network timeouts occur, and downstream processes expect specific schemas. Which error handling and validation architecture BEST ensures pipeline reliability and debuggability?

You need to join two large pandas DataFrames: df1 (100M rows) with columns [user_id, timestamp, event] and df2 (50K rows) with [user_id, category, segment]. The join is on user_id. After joining, 40% of rows have NaN in the category column because those user_ids don't exist in df2. You need to perform subsequent groupby operations on category. Which approach provides the BEST performance?

You're building a classification model for a rare disease diagnosis (0.1% prevalence). Your model achieves 99.5% accuracy, 15% precision, and 70% recall on the test set. The medical team needs to minimize false negatives (missing actual cases) while managing the workload of investigating positive predictions. What model adjustment strategy is MOST appropriate?

You've trained a random forest model with 100 trees for customer behavior prediction. Feature importance analysis shows that 'customer_id' has the highest importance score, followed by meaningful features like purchase_history and engagement_score. Cross-validation accuracy is 0.94, but when deployed to production with new customers (not in training), accuracy drops to 0.52. What is the MOST likely cause and solution?