IBM A1000-120 - Assessment: Data Science Foundations Advanced Practice Exam: Hard Questions 2025

A data science team is analyzing customer churn data with 95% of customers retained and 5% churned. After training a logistic regression model, it achieves 95% accuracy but fails to identify most churning customers. The business requires identifying at least 70% of potential churners even if it means more false positives. Which combination of techniques would BEST address this problem?

During exploratory data analysis of a manufacturing dataset with sensor readings, you observe that temperature measurements show a bimodal distribution with peaks at 72°F and 180°F, high kurtosis, and the mean is significantly different from the median. Which statistical approach would be MOST appropriate for analyzing the central tendency and why?

A retail analytics team needs to visualize the relationship between customer lifetime value (continuous), customer segment (5 categories), purchase frequency (continuous), and satisfaction score (ordinal 1-5) simultaneously for 10,000 customers. Which visualization strategy would MOST effectively communicate these multidimensional relationships to business stakeholders?

A pharmaceutical company is conducting hypothesis testing to determine if a new drug reduces blood pressure more effectively than a placebo. With a sample size of 50 patients per group, they observe a p-value of 0.048 (α = 0.05), a mean difference of 3.2 mmHg, and a 95% confidence interval of [0.1, 6.3] mmHg. The clinical significance threshold is 5 mmHg. What is the MOST appropriate interpretation and recommendation?

A data scientist is building a predictive model for fraud detection where false negatives (missing fraud) cost the company $5,000 per incident on average, while false positives (flagging legitimate transactions) cost $50 in investigation time. The current model has 85% precision and 75% recall on a test set with 2% fraud rate. Which approach would BEST optimize the model for business value?

During feature engineering for a time-series forecasting model predicting monthly sales, a data scientist creates lag features (sales from 1, 2, and 3 months prior) and rolling window statistics. When evaluating the model using standard k-fold cross-validation, it achieves excellent performance (R² = 0.92), but performs poorly in production. What is the MOST likely cause and appropriate solution?

A dataset contains customer income data with 15% missing values that are Missing Not At Random (MNAR) - specifically, high-income individuals are less likely to report income. The data scientist needs to prepare this feature for a credit risk model. Which imputation strategy would be MOST appropriate?

A multinational company wants to build a recommendation system using collaborative filtering on user-product interaction data from 50 countries. Initial analysis shows that product preferences vary dramatically by region, and the user-item matrix is 99.8% sparse. Which architectural approach would MOST effectively balance model performance and computational efficiency?

When comparing two machine learning models for production deployment, Model A (Random Forest) has training accuracy of 94%, validation accuracy of 89%, and test accuracy of 88%. Model B (Gradient Boosting) has training accuracy of 98%, validation accuracy of 87%, and test accuracy of 87%. Both models show stable performance across 5 different random seeds. Which assessment is MOST accurate?

A data science team is analyzing the relationship between advertising spend (X) and sales revenue (Y) using linear regression. They obtain R² = 0.76, but residual plots show a clear funnel pattern (heteroscedasticity) with variance increasing as spend increases. The Durbin-Watson statistic is 2.1. Which approach would MOST appropriately address the regression assumption violations?