Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Generative AI Leader practice bank has 2,191 more like it. Try 10 free: instant score, no account.
An enterprise is preparing customer transaction datasets to train a machine learning and generative AI model for personalized marketing across global markets. During a preliminary data quality assessment, the team identifies three primary issues:
"Standard", "standard", and "Standrd"), causing categories to fragment.Which strategy best resolves these issues across the core dimensions of data quality to ensure accurate, consistent, and representative training data?
Data quality management in AI and machine learning is the systematic process of evaluating, cleaning, and structuring datasets to ensure that training inputs are complete, consistent, and representative of real-world production environments. High-quality training data directly determines a model's predictive accuracy, fairness, and generalization capabilities.
NULL in BigQuery) prevents misinterpretation and retains the maximum signal from each training example."Standard" and "standard" are treated as a single unified entity rather than separate, fragmented classes that degrade model quality.This approach directly targets the fundamental dimensions of data quality—completeness, consistency, and relevance—without taking destructive shortcuts such as discarding entire feature sets or introducing synthetic noise.
This strategy involves including future customer purchase or conversion information as predictive features to compensate for missing data, removing whitespace delimiters in multi-word text strings, and bypassing dataset validation splits.
"red green blue" to "red/green/blue" or single continuous strings) prevents tokenizers from isolating individual semantic words and deriving actionable training signal.To ensure robust AI implementations, input features must be restricted strictly to information available in the exact same format at serving time, while preserving natural token boundaries in text data.
This approach suggests imputing all missing entries with placeholder numbers such as 0 or special characters, maintaining misspelled text as distinct categorical classes, and solving the geographic representation gap by narrowing business deployment exclusively to the domestic region.
0 or special characters to represent nulls causes machine learning systems to interpret those placeholders as legitimate measurements or extreme numerical signals, significantly corrupting model quality.Properly representing null values (such as using empty strings for CSV imports or explicit database NULL types) allows AI algorithms to handle missingness correctly without skewing feature distributions.
This option suggests using dimensionality reduction (such as Principal Component Analysis) as an automated tool to discard incomplete columns, keeping unstandardized categorical text under the belief that variations increase vocabulary diversity, and mathematically oversampling a single domestic region to compensate for international data gaps.
"Standard" vs "standrd") causes AI platforms and algorithms to interpret identical conceptual categories as entirely distinct classes.While dimensionality reduction is valuable in machine learning workflows, applying it to bypass essential data cleaning leads to poor model interpretability and degraded real-world performance.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated