Free PMI-CPMAI sample questions

Correct answers: Apache Kafka; Apache Airflow

Apache Kafka is a distributed streaming platform designed to handle high-throughput, real-time data feeds with excellent fault tolerance, making it ideal for ingesting IoT data.

Apache Airflow is a workflow orchestration tool that allows for programmatic authoring, scheduling, and monitoring of complex data pipelines, including both batch and streaming jobs. It satisfies the need to manage complex workflows.

Correct answer: False

This statement is false. The 'Seven Patterns of AI' (e.g., recognition, prediction, conversation) are high-level problem archetypes. They are most critically applied during Phase I (Business Understanding) to frame the business problem in AI terms and determine the general approach, long before specific algorithm selection occurs in Phase IV.

Correct answer: The model will fail to identify most of the employees who are actually at risk of leaving, defeating the project's purpose.

Recall (or sensitivity) measures the model's ability to find all the relevant cases within a dataset. In this scenario, low recall for the 'attrition' class means the model is failing to identify actual at-risk employees (high false negatives). High accuracy is misleading due to class imbalance (most employees don't leave). The primary business risk is that the company will miss the opportunity to retain valuable talent because the model fails to flag them. Option A describes low precision (high false positives).

Correct answer: F-beta Score with beta < 1

The F-beta score is a generalization of the F1-score that allows for weighting precision over recall, or vice versa. A beta value less than 1 (e.g., F0.5-score) gives more weight to precision (minimizing False Positives), while a beta greater than 1 gives more weight to recall (minimizing False Negatives). Since a default (False Positive) is five times more costly, the model should prioritize precision. Therefore, an F-beta score with beta < 1 is the most appropriate metric. The F1-score gives equal weight to both. AUC is a general measure of separability and does not directly account for asymmetric costs.

Correct answer: Data Lineage

Data lineage is the process of understanding, recording, and visualizing data as it flows from data sources to consumption. This includes all transformations the data underwent along the way. Implementing robust data lineage is essential for meeting the requirements of traceability, auditing, and debugging AI models. Data masking and anonymization are privacy techniques, and data augmentation is a method for increasing training data size.

Correct answer: To centralize the storage, retrieval, and management of curated features for both model training and real-time inference.

A feature store is a central repository for features used in machine learning. Its primary purpose is to solve the problem of feature consistency between training and serving, reduce duplicate feature engineering work across teams, and provide a reliable, low-latency source of feature data for real-time inference. This centralization is a key component of a mature MLOps strategy.

Correct answer: Use a pre-trained foundational LLM and apply a Retrieval-Augmented Generation (RAG) approach using product documentation.

With limited specific data, training an LLM from scratch is infeasible. The most effective strategy is to leverage a powerful pre-trained model and augment its knowledge with domain-specific information. Retrieval-Augmented Generation (RAG) is designed for this; it retrieves relevant information from a knowledge base (like product documentation) and provides it to the LLM as context to generate an accurate answer. This is faster, cheaper, and more effective than fine-tuning with insufficient data or building from scratch.

Correct answer: SHAP is a technique that assigns an importance value to each feature for an individual prediction, showing how much each factor contributed to the model's output.

This is the most accurate description. SHAP (SHapley Additive exPlanations) is a game theory-based approach to explain the output of any machine learning model. It connects optimal credit allocation with local explanations using the classic Shapley values. For a specific prediction, it shows which features pushed the prediction higher or lower and by how much, providing crucial local interpretability for 'black box' models.

Correct answers: Concept Drift; Data Drift

Concept drift occurs when the statistical properties of the target variable change over time. For churn, this could mean the underlying reasons why customers leave have changed (e.g., a new competitor entered the market, changing customer behavior).

Data drift (or feature drift) happens when the statistical properties of the input features change. For example, the average customer tenure or product usage patterns might have shifted since the model was trained, making the production data different from the training data.

Correct answer: A cross-functional team with roles like Data Engineer, ML Scientist, MLOps Engineer, and a Product Owner who represents the business stakeholders.

AI projects are inherently multi-disciplinary. A cross-functional team structure that brings together all necessary skills (data engineering, machine learning science, operations/deployment, and business context) fosters collaboration, reduces hand-off friction, and ensures the end product is aligned with business needs and technically robust. Siloed approaches (A) are inefficient and prone to error. A team of only data scientists (B) would lack critical engineering and business skills. Outsourcing core development (D) can work but is not inherently the most successful structure.