WGU D429 Objective Assessment V1 & V2
Exam| D429 Introduction To AI For
Computer Scientists OA Exam | All
Questions And Solutions (Rationales) |
Exam Prep Q&A 2026/2027
1. A software development team is designing an AI system to classify customer support
messages into categories such as billing, technical issues, and account access. Which
approach is most appropriate for this task?
A. Supervised learning using labeled examples of customer messages
B. Unsupervised clustering without evaluating the resulting groups
C. Reinforcement learning based exclusively on delayed financial rewards
D. Manual rule execution without analyzing historical examples
Answer: A
Rationale: Supervised learning trains a model using labeled examples. Each message is paired
with its intended category, allowing the model to learn patterns that help classify previously
unseen messages.
2. A computer scientist is explaining the difference between artificial intelligence and
traditional rule-based programming to a project manager. Which statement best describes
machine learning?
A. It requires programmers to specify every possible input-output relationship.
B. It enables systems to learn patterns from data and use them to make predictions or decisions.
C. It eliminates the need for data quality controls.
D. It guarantees correct decisions whenever sufficient computing power is available.
Answer: B
,Rationale: Machine learning algorithms infer patterns from data rather than relying
exclusively on manually written rules. Their performance still depends on factors such as data
quality, model selection, and evaluation.
3. An organization trains a model to predict whether customers will cancel subscriptions.
The dataset contains historical customer activity and a column indicating whether each
customer eventually canceled. What role does the cancellation column play during
supervised training?
A. It is always an unsupervised clustering variable.
B. It represents the computational complexity of the model.
C. It serves as the target label the model learns to predict.
D. It must be removed before the training process begins.
Answer: C
Rationale: The target label represents the outcome the model is intended to predict. Historical
cancellation status supplies the supervision needed to learn relationships between customer
features and cancellation outcomes.
4. A data scientist develops a model that performs exceptionally well on training data but
makes unreliable predictions on new customer records. Which problem is most likely
occurring?
A. Underfitting caused by a model that cannot learn training patterns
B. Data normalization that always improves generalization
C. Successful generalization across all possible data distributions
D. Overfitting to patterns specific to the training dataset
Answer: D
Rationale: Overfitting occurs when a model learns training-specific details, including noise,
rather than generalizable relationships. Evaluation on unseen data helps reveal this problem.
5. A researcher wants to estimate how accurately an AI model will perform on previously
unseen examples. Which practice provides the most direct evidence of generalization?
A. Evaluating the finalized model on a representative test dataset that was not used for training
or model selection
,B. Repeatedly evaluating the model on its training dataset
C. Selecting the model with the largest number of parameters
D. Reporting only the highest accuracy observed during training
Answer: A
Rationale: A separate test dataset provides an estimate of performance on unseen examples.
It should remain isolated from training and model selection to reduce the risk of optimistic
evaluation.
6. A bank is developing an AI system to identify potentially fraudulent transactions.
Missing fraudulent transactions are especially costly, although false alarms also require
investigation. Which evaluation approach is most appropriate?
A. Measure training time and ignore classification errors.
B. Examine recall, precision, and the costs associated with false positives and false negatives.
C. Select the model with the highest number of input features.
D. Optimize accuracy alone without examining the class distribution.
Answer: B
Rationale: Recall measures the proportion of actual fraud cases detected, while precision
measures the proportion of flagged transactions that are fraudulent. Evaluating both
alongside operational costs supports an appropriate decision threshold.
7. A machine learning engineer discovers that the target variable was accidentally included
among the input features during training. The resulting model reports nearly perfect test
accuracy. What is the primary concern?
A. The model has necessarily learned a more general representation.
B. The model must have excessive regularization.
C. Target leakage has allowed information about the answer to enter the prediction process.
D. The model cannot calculate probabilities.
Answer: C
, Rationale: Target leakage occurs when information unavailable at the intended prediction
time reveals or strongly implies the target. It can produce misleading performance estimates
and poor real-world results.
8. A company has millions of customer records but no predefined categories. It wants to
discover groups of customers with similar purchasing behaviors. Which method best fits
this objective?
A. Binary classification with known labels
B. Regression using a continuous target
C. Supervised classification with fabricated labels
D. Unsupervised clustering based on customer features
Answer: D
Rationale: Clustering is an unsupervised learning technique that groups observations
according to similarities in their features. The groups can subsequently be interpreted and
validated for business usefulness.
9. A model predicts house prices using property size, location, and age. Which type of
machine learning problem is this?
A. Regression
B. Classification
C. Clustering
D. Association rule mining
Answer: A
Rationale: Regression predicts a numerical quantity, such as a house price. Classification
instead predicts a discrete class, while clustering identifies groups without requiring
predefined labels.
10. A team is building a model to identify handwritten digits from images. Which output
configuration is most appropriate when each image represents exactly one digit from zero
through nine?
A. A single unrestricted text response with no evaluation criteria
B. A ten-class classification output, typically producing class scores or probabilities
Exam| D429 Introduction To AI For
Computer Scientists OA Exam | All
Questions And Solutions (Rationales) |
Exam Prep Q&A 2026/2027
1. A software development team is designing an AI system to classify customer support
messages into categories such as billing, technical issues, and account access. Which
approach is most appropriate for this task?
A. Supervised learning using labeled examples of customer messages
B. Unsupervised clustering without evaluating the resulting groups
C. Reinforcement learning based exclusively on delayed financial rewards
D. Manual rule execution without analyzing historical examples
Answer: A
Rationale: Supervised learning trains a model using labeled examples. Each message is paired
with its intended category, allowing the model to learn patterns that help classify previously
unseen messages.
2. A computer scientist is explaining the difference between artificial intelligence and
traditional rule-based programming to a project manager. Which statement best describes
machine learning?
A. It requires programmers to specify every possible input-output relationship.
B. It enables systems to learn patterns from data and use them to make predictions or decisions.
C. It eliminates the need for data quality controls.
D. It guarantees correct decisions whenever sufficient computing power is available.
Answer: B
,Rationale: Machine learning algorithms infer patterns from data rather than relying
exclusively on manually written rules. Their performance still depends on factors such as data
quality, model selection, and evaluation.
3. An organization trains a model to predict whether customers will cancel subscriptions.
The dataset contains historical customer activity and a column indicating whether each
customer eventually canceled. What role does the cancellation column play during
supervised training?
A. It is always an unsupervised clustering variable.
B. It represents the computational complexity of the model.
C. It serves as the target label the model learns to predict.
D. It must be removed before the training process begins.
Answer: C
Rationale: The target label represents the outcome the model is intended to predict. Historical
cancellation status supplies the supervision needed to learn relationships between customer
features and cancellation outcomes.
4. A data scientist develops a model that performs exceptionally well on training data but
makes unreliable predictions on new customer records. Which problem is most likely
occurring?
A. Underfitting caused by a model that cannot learn training patterns
B. Data normalization that always improves generalization
C. Successful generalization across all possible data distributions
D. Overfitting to patterns specific to the training dataset
Answer: D
Rationale: Overfitting occurs when a model learns training-specific details, including noise,
rather than generalizable relationships. Evaluation on unseen data helps reveal this problem.
5. A researcher wants to estimate how accurately an AI model will perform on previously
unseen examples. Which practice provides the most direct evidence of generalization?
A. Evaluating the finalized model on a representative test dataset that was not used for training
or model selection
,B. Repeatedly evaluating the model on its training dataset
C. Selecting the model with the largest number of parameters
D. Reporting only the highest accuracy observed during training
Answer: A
Rationale: A separate test dataset provides an estimate of performance on unseen examples.
It should remain isolated from training and model selection to reduce the risk of optimistic
evaluation.
6. A bank is developing an AI system to identify potentially fraudulent transactions.
Missing fraudulent transactions are especially costly, although false alarms also require
investigation. Which evaluation approach is most appropriate?
A. Measure training time and ignore classification errors.
B. Examine recall, precision, and the costs associated with false positives and false negatives.
C. Select the model with the highest number of input features.
D. Optimize accuracy alone without examining the class distribution.
Answer: B
Rationale: Recall measures the proportion of actual fraud cases detected, while precision
measures the proportion of flagged transactions that are fraudulent. Evaluating both
alongside operational costs supports an appropriate decision threshold.
7. A machine learning engineer discovers that the target variable was accidentally included
among the input features during training. The resulting model reports nearly perfect test
accuracy. What is the primary concern?
A. The model has necessarily learned a more general representation.
B. The model must have excessive regularization.
C. Target leakage has allowed information about the answer to enter the prediction process.
D. The model cannot calculate probabilities.
Answer: C
, Rationale: Target leakage occurs when information unavailable at the intended prediction
time reveals or strongly implies the target. It can produce misleading performance estimates
and poor real-world results.
8. A company has millions of customer records but no predefined categories. It wants to
discover groups of customers with similar purchasing behaviors. Which method best fits
this objective?
A. Binary classification with known labels
B. Regression using a continuous target
C. Supervised classification with fabricated labels
D. Unsupervised clustering based on customer features
Answer: D
Rationale: Clustering is an unsupervised learning technique that groups observations
according to similarities in their features. The groups can subsequently be interpreted and
validated for business usefulness.
9. A model predicts house prices using property size, location, and age. Which type of
machine learning problem is this?
A. Regression
B. Classification
C. Clustering
D. Association rule mining
Answer: A
Rationale: Regression predicts a numerical quantity, such as a house price. Classification
instead predicts a discrete class, while clustering identifies groups without requiring
predefined labels.
10. A team is building a model to identify handwritten digits from images. Which output
configuration is most appropriate when each image represents exactly one digit from zero
through nine?
A. A single unrestricted text response with no evaluation criteria
B. A ten-class classification output, typically producing class scores or probabilities
