comptia data+ practice test Advanced Practice Exam: Hard Questions 2025

A healthcare organization is building an analytics platform that ingests EHR data from multiple clinics. Analysts need near-real-time dashboards, while data scientists need reproducible historical datasets. The current design uses a single OLTP database as both the operational store and analytics source, causing lock contention and inconsistent query results during peak hours. Which architecture change BEST addresses both workload isolation and reproducibility requirements?

A retailer runs a clustering model to segment customers using features including total spend, number of transactions, and average basket size. The model produces unstable clusters across runs and the business reports that segments shift dramatically week-to-week even when behavior appears similar. Investigation shows total spend has a heavy-tailed distribution with extreme outliers. Which change is MOST likely to produce more stable, meaningful clusters without discarding high-value customers?

A bank is developing a churn prediction model. The dataset includes 2 years of monthly snapshots per customer. An analyst proposes random train/test splitting across all rows to maximize sample size. You notice unusually high AUC during validation but poor performance after deployment. Which evaluation approach MOST directly prevents the likely issue while reflecting real-world scoring?

A manufacturer uses an anomaly detection model on sensor data to flag potential equipment failures. Failures are rare, and missing a true failure is very costly, but excessive false positives cause production stoppages. The model outputs calibrated probabilities. Which approach BEST aligns model thresholding with business risk while remaining defensible to stakeholders?

A data analyst is asked to quantify the impact of a new recommendation widget on average order value (AOV). The company rolled it out first to high-traffic markets and later to low-traffic markets. A simple pre/post comparison shows a large uplift, but executives suspect confounding because seasonality and market differences are significant. Which analysis method is MOST appropriate to isolate the widget’s effect given the rollout pattern?

A team aggregates clickstream events into daily active users (DAU). After a pipeline refactor, DAU increases by 18% overnight. Investigation shows some users have multiple device identifiers that sometimes map to a single account, and late-arriving events can arrive up to 48 hours late. Which troubleshooting step is MOST likely to pinpoint the root cause of the jump?

A dashboard shows conversion rate by marketing channel. For several channels, conversion rates fluctuate wildly day-to-day because volumes are low. Stakeholders demand a visualization that preserves the ability to compare channels while reducing noise and avoiding misleading rankings. Which visualization approach is BEST?

A finance team reports that a new dashboard is 'lying' because totals differ between two charts: one shows monthly revenue by region; another shows monthly revenue by product line. Both are sourced from the same dataset. You discover the underlying fact table contains one row per invoice line item and includes both region and product, but some invoices have multiple regions due to split fulfillment, creating duplicate revenue attribution when joined to the region dimension. What is the BEST remediation to ensure consistent, explainable totals across visualizations?

A public-sector agency must publish an open dataset containing incident records. The dataset includes quasi-identifiers (ZIP code, age, incident date) and a sensitive attribute (incident type). Leadership wants to minimize re-identification risk while keeping the dataset useful for trend analysis. Which approach BEST balances privacy and utility?

A data platform ingests vendor files daily. The vendor occasionally changes column meaning without renaming fields (e.g., 'status' values expand, or a numeric field changes from dollars to cents). Downstream models silently degrade before anyone notices. The team already has basic schema checks for column presence and data types. What is the MOST effective additional control to detect these changes early?