View Full PMI CPMAI Exam Dumps and Practice Test Dumps.

 

Question 121

Which factor is most important when determining whether an AI use case should move from an idea to formal project evaluation?

1. The availability of a popular AI framework
2. The number of employees interested in AI
3. The potential business value and feasibility of the use case
4. The amount of publicity surrounding similar AI systems

Correct Answer: 3

Explanation

An AI use case should be evaluated based on both its potential value and its feasibility. Business value may include improved efficiency, reduced costs, better decision-making, increased revenue, or improved customer experiences. Feasibility considers whether the organization has suitable data, technology, skills, infrastructure, funding, and operational support. A popular framework or high employee interest does not necessarily indicate that an AI initiative is worthwhile. Similarly, publicity around another organization’s implementation does not establish suitability for a particular organization. Evaluating value and feasibility together helps project teams identify opportunities that are meaningful, realistic, and aligned with organizational priorities.

Question 122

What is the primary purpose of a risk register in an AI project?

1. To record identified risks, their characteristics, owners, and planned responses
2. To replace the project schedule
3. To document only technical defects
4. To automatically eliminate project risks

Correct Answer: 1

Explanation

A risk register is a structured record used to identify, analyze, monitor, and manage project risks. For an AI project, entries may include data risks, privacy concerns, security threats, model limitations, operational dependencies, regulatory concerns, and vendor risks. A risk register can document each risk’s probability, potential impact, owner, response strategy, status, and related actions. It does not replace the project schedule or automatically eliminate risks. Instead, it gives the project team visibility into uncertainty and provides a mechanism for tracking responses over time. Maintaining the register throughout the lifecycle supports proactive rather than purely reactive risk management.

Question 123

Which data quality characteristic describes whether values follow the required format, rules, or constraints?

1. Timeliness
2. Validity
3. Completeness
4. Uniqueness

Correct Answer: 2

Explanation

Validity describes whether data conforms to defined formats, rules, ranges, or constraints. For example, a date field should contain valid dates, a percentage should fall within an appropriate range, and a required category should use an accepted value. Invalid data can create errors during analysis, training, or model inference. Completeness focuses on whether required information is present, timeliness concerns how current the information is, and uniqueness addresses duplicate records. Validity checks are therefore an important component of data-quality assessment. Identifying invalid values early allows teams to correct, transform, exclude, or otherwise appropriately handle problematic records before they affect an AI system.

Question 124

Why is a representative training dataset important for an AI model?

1. It guarantees zero bias
2. It eliminates the need for validation
3. It guarantees the model will never drift
4. It helps the model learn patterns that reflect the intended population and use case

Correct Answer: 4

Explanation

A representative dataset helps an AI model learn patterns that reasonably reflect the population, situations, and conditions in which the system is expected to operate. If important groups or scenarios are poorly represented, the model may perform differently when encountering those cases in production. Representativeness does not guarantee that a model will be unbiased or eliminate the need for validation and monitoring. Instead, it provides a stronger foundation for generalization and meaningful evaluation. Project teams should examine how data was collected, which populations are included, which cases may be missing, and whether the dataset reflects the intended operational environment.

Question 125

Which technique is commonly used to reduce the number of variables while preserving important information in a dataset?

1. Data duplication
2. Principal component analysis
3. Data labeling
4. Access control

Correct Answer: 2

Explanation

Principal component analysis, or PCA, is a dimensionality-reduction technique that transforms correlated variables into a smaller set of components that capture substantial variation in the original data. Reducing dimensionality can simplify datasets, decrease computational requirements, and sometimes help address problems associated with highly correlated features. PCA does not perform data labeling or access control, and duplicating data does not reduce dimensionality. However, PCA results can be less directly interpretable than the original variables, so teams should consider whether interpretability is important for the intended application. The technique should therefore be selected based on the project’s technical and business requirements.

Question 126

A model performs extremely well on training data but poorly on unseen data. What is the most likely problem?

1. Overfitting
2. Underutilization of storage
3. Data encryption
4. Excessive monitoring

Correct Answer: 1

Explanation

Overfitting occurs when a model learns the training data too closely, including patterns or noise that do not generalize to unseen examples. As a result, the model may achieve excellent training performance while producing substantially weaker results on validation or test data. Techniques such as regularization, simpler models, cross-validation, early stopping, feature selection, and obtaining more representative training data can help reduce overfitting depending on the situation. Evaluating performance only on training data can conceal the problem. Comparing training performance with validation or test performance provides evidence about how well the model generalizes beyond the examples it used during learning.

Question 127

Which metric is particularly useful when the cost of false negatives is significantly higher than the cost of false positives?

1. Accuracy
2. Precision
3. Recall
4. R-squared

Correct Answer: 3

Explanation

Recall measures the proportion of actual positive cases that the model correctly identifies. It is particularly important when missing a positive case creates significant consequences. For example, in a system designed to identify potentially fraudulent transactions or safety-related events, failing to identify a true case may be more costly than investigating some additional false alarms. Increasing recall can sometimes reduce precision, so the appropriate balance depends on the business context and risk tolerance. Accuracy alone may be misleading when classes are imbalanced. Project teams should therefore select metrics according to the consequences of different types of prediction errors.

Question 128

What is the primary purpose of cross-validation during model development?

1. To permanently increase the training dataset
2. To estimate how a model is likely to perform on unseen data
3. To eliminate the need for a final test set in every situation
4. To guarantee that the model is unbiased

Correct Answer: 2

Explanation

Cross-validation divides available training data into multiple subsets and repeatedly trains and evaluates the model using different portions for training and validation. This provides a more robust estimate of how the model may perform on unseen data and can help compare model configurations. Cross-validation does not guarantee that a model is unbiased, and it does not permanently increase the amount of available data. Depending on the project design, a separate test dataset may still be retained for final evaluation. Cross-validation is particularly useful when the available dataset is limited because it allows multiple validation arrangements while maintaining a structured evaluation process.

Question 129

Which activity is most appropriate for determining whether an AI solution can be supported after deployment?

1. Operational readiness assessment
2. Increasing model complexity
3. Removing monitoring controls
4. Reducing documentation

Correct Answer: 1

Explanation

An operational readiness assessment determines whether the organization is prepared to operate and support an AI system in its intended environment. Areas may include infrastructure, monitoring, support processes, security controls, user training, incident response, model ownership, data pipelines, deployment procedures, and rollback capabilities. A technically successful model may still fail operationally if the organization cannot maintain or monitor it effectively. Increasing model complexity does not address operational readiness, while removing controls and documentation can increase risk. Assessing readiness before production deployment helps identify gaps that could affect reliability, maintainability, security, user adoption, and the organization’s ability to respond to problems.

Question 130

What is the main purpose of defining acceptance criteria for an AI project?

1. To determine who writes the source code
2. To specify measurable conditions that must be satisfied for an outcome to be accepted
3. To prevent stakeholders from reviewing the solution
4. To guarantee a specific algorithm is used

Correct Answer: 2

Explanation

Acceptance criteria define the measurable conditions that must be satisfied for a deliverable or AI solution to be considered acceptable. Criteria may cover model performance, business outcomes, usability, security, privacy, response time, reliability, or other requirements relevant to the project. Clear acceptance criteria help stakeholders and project teams establish a shared understanding of success before implementation is completed. They also provide a basis for testing and validation. Acceptance criteria do not prescribe a specific algorithm or prevent stakeholder involvement. Instead, they connect project requirements to objective conditions that can be evaluated before a solution is formally accepted.

Question 131

Which approach is most appropriate for managing a change that could significantly affect the approved scope of an AI project?

1. Implement it immediately without analysis
2. Ignore the requested change
3. Evaluate the change through an established change-control process
4. Allow only developers to approve it

Correct Answer: 3

Explanation

Significant changes to an AI project should normally be evaluated through an established change-control process. The assessment can consider impacts on scope, schedule, budget, resources, architecture, data, risks, compliance, quality, and expected benefits. Appropriate stakeholders or governance authorities can then decide whether the change should be approved, rejected, modified, or deferred. Implementing major changes without analysis can create uncontrolled scope expansion and introduce new risks. Ignoring all changes is also inappropriate because legitimate requirements may evolve. Change control provides a structured way to manage necessary changes while maintaining visibility and alignment with the project’s approved objectives.

Question 132

What is the primary purpose of data profiling before developing an AI model?

1. To understand the structure, quality, distributions, and characteristics of the available data
2. To automatically select the final algorithm
3. To guarantee regulatory compliance
4. To deploy the model into production

Correct Answer: 1

Explanation

Data profiling examines the structure, contents, distributions, relationships, and quality characteristics of a dataset. It can reveal missing values, duplicates, unusual values, inconsistent formats, outliers, class imbalance, and other issues that may affect AI development. Profiling helps the project team understand what data is actually available before selecting preprocessing techniques or modeling approaches. It does not automatically select an algorithm or guarantee compliance. Deployment is also a separate lifecycle activity. By identifying data characteristics and potential problems early, profiling helps teams make better decisions about data preparation, feature engineering, sampling, validation, and whether additional data acquisition may be necessary.

Question 133

Which statement best describes the purpose of a model card or similar model documentation?

1. It replaces all testing activities
2. It provides information about the model’s purpose, performance, limitations, and relevant considerations
3. It guarantees that the model will remain accurate indefinitely
4. It prevents users from accessing the model

Correct Answer: 2

Explanation

Model documentation provides structured information about an AI model so that stakeholders can understand how and where it should be used. Depending on the documentation approach, it may describe the model’s intended purpose, training information, evaluation results, limitations, known risks, applicable conditions, and performance across relevant groups. This information supports transparency, governance, responsible use, and informed decision-making. Documentation does not replace testing or guarantee future accuracy because model behavior can change after deployment. It also does not inherently prevent access. Instead, it provides users, reviewers, and operators with important context for understanding the model and recognizing situations in which its use may be inappropriate.

Question 134

Which action best helps prevent data leakage during supervised model development?

1. Using information that would only be available after the prediction event
2. Allowing future outcome information into the training features
3. Ensuring features are derived only from information legitimately available at prediction time
4. Combining training and test records without controls

Correct Answer: 3

Explanation

Data leakage occurs when information that would not legitimately be available at prediction time becomes available to the model during training or evaluation. This can make performance appear substantially better than it will be in real-world use. To prevent leakage, teams should ensure that features are derived only from information that would genuinely be available when the prediction is made. Temporal relationships, preprocessing procedures, target-derived variables, and dataset splits should be carefully reviewed. Allowing future information into training features can produce misleading results. Preventing leakage is essential because a model with artificially inflated evaluation performance may fail significantly after deployment.

Question 135

What is a key benefit of maintaining version control for AI project artifacts?

1. It allows teams to track changes and restore or reproduce previous versions
2. It guarantees that every model is unbiased
3. It eliminates the need for documentation
4. It prevents all unauthorized access automatically

Correct Answer: 1

Explanation

Version control allows teams to track changes to important project artifacts such as source code, configuration files, model definitions, documentation, and sometimes data or data references. It helps identify what changed, when it changed, and who made the change. Teams can use this history to reproduce previous results, investigate problems, compare versions, and restore earlier configurations when necessary. Version control does not guarantee fairness or eliminate the need for documentation and security controls. It is one component of a broader governance and development process. Proper version management becomes particularly valuable when multiple contributors work on an AI project concurrently.

Question 136

Which situation best illustrates the use of human-in-the-loop AI?

1. A model makes every decision without review
2. A human reviews or approves model recommendations before consequential decisions are finalized
3. A model is permanently disconnected from users
4. A system automatically deletes every uncertain prediction

Correct Answer: 2

Explanation

Human-in-the-loop systems incorporate human involvement into the decision process, particularly when model outputs may have significant consequences or require contextual judgment. For example, an AI system may recommend a decision while an authorized employee reviews the recommendation and approves, rejects, or modifies it before the final action occurs. The exact level of human involvement depends on the use case and risk. Human-in-the-loop does not mean that humans must perform every task manually. Instead, AI can assist with analysis while humans retain an appropriate decision or review role. This approach can provide additional oversight when automated predictions require contextual interpretation.

Question 137

Why is a rollback plan important before deploying a new AI model?

1. It guarantees that deployment will succeed
2. It increases the model’s training accuracy
3. It provides a way to return to a previously stable version if serious problems occur
4. It eliminates the need for production monitoring

Correct Answer: 3

Explanation

A rollback plan provides a defined procedure for returning an AI system to a previously stable model or configuration if the new deployment causes unacceptable problems. Issues may include unexpected performance degradation, security concerns, data compatibility problems, excessive latency, or harmful business outcomes. Having a rollback mechanism reduces the time required to respond to serious incidents and can limit operational impact. A rollback plan does not guarantee successful deployment or improve model training accuracy. Monitoring remains necessary because teams need evidence that a deployment is behaving as expected. Rollback is therefore an important component of responsible deployment and operational risk management.

Question 138

Which measure is most appropriate for evaluating the average absolute difference between predicted and actual values in a regression model?

1. Recall
2. Precision
3. F1 score
4. Mean absolute error

Correct Answer: 4

Explanation

Mean absolute error, or MAE, measures the average absolute difference between predicted values and actual values in a regression problem. It provides an intuitive indication of the typical magnitude of prediction errors because it expresses errors in the same units as the target variable. Unlike classification metrics such as precision, recall, and F1 score, MAE is designed for continuous-value predictions. MAE treats errors linearly, meaning larger errors contribute proportionally more to the total. The appropriate regression metric should still depend on the business context and consequences of errors. Teams may also compare MAE with other metrics such as RMSE or R-squared.

Question 139

Which activity can help an organization evaluate whether an AI initiative is delivering the expected business benefits after implementation?

1. Benefits realization monitoring
2. Increasing the number of model parameters
3. Removing business metrics
4. Avoiding stakeholder feedback

Correct Answer: 1

Explanation

Benefits realization monitoring evaluates whether the AI initiative is producing the business outcomes that justified the project. Depending on the use case, these outcomes may include reduced processing time, lower operating costs, increased revenue, improved service quality, fewer errors, or other defined key performance indicators. Technical model performance alone does not prove that an AI initiative is delivering business value. Monitoring business outcomes allows project and business teams to compare actual results with the expected benefits established during planning. If benefits are not being achieved, the organization can investigate adoption, process integration, model performance, data quality, or other factors contributing to the gap.

Question 140

Which practice best supports secure access to an AI model and its associated artifacts?

1. Giving every project member administrator privileges
2. Sharing credentials among team members
3. Applying role-based access controls according to responsibilities
4. Storing credentials in publicly accessible documentation

Correct Answer: 3

Explanation

Role-based access control limits access according to the responsibilities and permissions required for a person’s role. In an AI environment, different users may need different levels of access to model artifacts, training data, deployment systems, configuration files, or monitoring tools. Applying least-privilege principles reduces unnecessary exposure and can limit the impact of compromised accounts. Sharing credentials or granting administrator access broadly increases security risk. Credentials should also be protected using appropriate security mechanisms rather than placed in public documentation. Access controls should be reviewed periodically to ensure that permissions remain appropriate as project roles and responsibilities change.