Practice C14-025–036
C14-025 — Ingest / Integrate / Quality · Foundational · Recall
What should be captured as data is acquired and ingested? - A. Metadata such as origin/source, size, currency, content/structure, plus profiling/classification and quality context. - B. Only the physical file name. - C. Only the predictive target label. - D. Only the final visualization title.
C14-026 — Ingest / Integrate / Quality · Standard · Understand
Why should Data Quality be assessed before integrating multiple sources? - A. Because integrated data is automatically too large to profile. - B. It reveals whether the sources are usable and how they must be aligned, preventing poor evidence from being embedded into integration and models. - C. Because DQ replaces Metadata once sources are combined. - D. Because all sources must be cleansed to perfect accuracy before any analysis.
C14-027 — Ingest / Integrate / Quality · Standard · Distinguish
Daily web behavior and monthly customer-value data share a customer key but are treated as if they describe the same time period. What is the best corrective action? - A. Switch from ELT to ETL. - B. Add more rows to the daily source. - C. Align and document temporal granularity/timing before analysis. - D. Use supervised learning instead of predictive analytics.
C14-028 — Ingest / Integrate / Quality · Difficult · Apply
Clickstream uses cookies, CRM uses customer IDs, and service systems use account IDs. A model needs one consistent customer view. Which supporting discipline is especially important? - A. Data visualization standards. - B. Sentiment analysis. - C. Volume management only. - D. Reference and Master Data / trusted identifiers and entity alignment.
C14-029 — Ingest / Integrate / Quality · Difficult · Distinguish
Which statement best distinguishes Metadata quality/context from Data Quality in a Big Data source? - A. Metadata tells what the source is, means, where it came from, and how it relates; Data Quality assesses whether its values are fit/trustworthy enough for use. - B. Metadata measures accuracy; Data Quality records lineage only. - C. Metadata is only technical schema; Data Quality is only business definitions. - D. They are synonyms in a data lake.
C14-030 — Ingest / Integrate / Quality · Expert · Distinguish
A team integrates five providers first, builds a model, and only then discovers that two feeds use incompatible definitions and one has severe missingness. What process error best explains the failure? - A. The team used too many data sources; Chapter 14 recommends only one source per model. - B. Source profiling/Data Quality and semantic alignment were postponed until after integration/modeling. - C. The model should have been prescriptive rather than predictive. - D. The solution needed a faster serving layer.
C14-031 — Modeling / Training / Visualization · Foundational · Recall
What is over-fitting? - A. A model that is too slow in production. - B. A model trained on too little historical data but still generalizes well. - C. A model learns training-specific patterns/noise too closely and performs poorly on unseen data. - D. A visualization with too many lines.
C14-032 — Modeling / Training / Visualization · Standard · Distinguish
Which data partition should remain independent for the final estimate of generalization after model choices are made? - A. Training set. - B. Validation set. - C. Metadata repository. - D. Test set.
C14-033 — Modeling / Training / Visualization · Difficult · Apply
A team checks the same 'test' set after every feature change and keeps the changes that improve its score. What is the main problem? - A. The test set is no longer independent; it has become part of model selection/tuning. - B. The test set has become a training set in the strict sense that model parameters were fit directly to it. - C. The model must now use unsupervised learning. - D. The team should add more visualization before evaluating accuracy.
C14-034 — Modeling / Training / Visualization · Difficult · Apply
A fraud team finds a small group of extreme transactions and proposes deleting them because they are statistical outliers. What is the best response? - A. Delete them automatically because all outliers reduce model quality. - B. Investigate the outliers before removal because they may be valid rare events and the fraud signal being sought. - C. Average them into normal transactions so the distribution is smoother. - D. Move them to the speed layer and ignore them during training.
C14-035 — Modeling / Training / Visualization · Standard · Understand
Which is the strongest principle for a Chapter 14 visualization? - A. Use the most complex interactive graphic available. - B. Remove explanatory text so the visual speaks for itself. - C. It should answer a defined question for a defined audience clearly and neutrally, with enough context for valid interpretation. - D. Always maximize the number of variables displayed.
C14-036 — Modeling / Training / Visualization · Expert · Apply
An executive chart truncates the axis and omits the relevant comparison group, making a small effect look dramatic. The underlying calculation is correct. What is the primary failure? - A. Data Quality, because any misleading chart proves the source values are wrong. - B. Metadata, because every chart is Metadata. - C. Volume, because the chart summarized too much data. - D. Visualization governance/interpretation validity: the presentation is misleading even though the calculation is correct.