All MCQs

A held-out test set provides an independent estimate of generalization performance. Using it during model development would make that estimate overly optimistic.

Integration must determine which records from different sources refer to the same real-world entity. Incorrect matches can create duplicates or combine unrelated customers.

Provenance supports traceability, reproducibility, and assessment of data quality or limitations. It helps analysts understand collection methods, ownership, and transformations.

A predefined metric makes model comparison consistent and reduces the temptation to select favorable results afterward. The metric should reflect the practical costs and benefits of the application.

Problem understanding establishes what the project must achieve and how results will be used. Without it, technically strong analysis may solve the wrong problem.

A data-quality profile systematically measures whether data meet expected standards and constraints. It helps prioritize cleaning and assess whether analysis is trustworthy.

Right skew means most observations are lower while a tail extends toward larger values. The mean is often pulled in the direction of that tail.

Nominal categories have no meaningful numerical order, so counts and proportions describe their distribution. A mean or regression slope is not inherently meaningful for the labels.

Outliers can result from data errors or represent important real phenomena. Their meaning should be assessed using context before applying any treatment.

Unintended duplicates give some cases extra weight and can inflate frequencies or apparent performance. They should be investigated before deciding whether to remove them.

When the probability of missingness depends on the missing value itself, observed cases may be systematically different. Simple deletion or imputation can then produce biased conclusions.

Correlation can arise from a shared factor such as hot weather rather than direct causation. Establishing causality requires stronger design or assumptions than a scatter plot provides.