AI-900 Practice Test 42 — 15 Questions

Question 1

A bank uses a machine learning model to predict credit card fraud. The model's output is a probability score. The business wants to minimize the number of false positives (legitimate transactions incorrectly flagged as fraud) because these cause customer dissatisfaction. At the same time, they must also catch most fraudulent transactions. Which metric should the bank optimize to balance these two goals?

Accepted Answer

D: F1 score. The F1 score is the harmonic mean of precision and recall, making it the ideal metric when a balance between minimizing false positives (precision) and catching most fraudulent transactions (recall) is required. In this credit card fraud detection scenario, optimizing F1 ensures the model reduces customer dissatisfaction from false positives while still maintaining high detection of actual fraud.

Answer

A: Accuracy

Answer

B: Precision

Answer

C: Recall

Question 2

What is the purpose of training data labeling in computer vision model development?

Accepted Answer

Adding ground-truth annotations to training images so the model learns what to predict. Training data labeling is the process of adding ground-truth annotations (e.g., bounding boxes, segmentation masks, or class labels) to each training image. This supervised learning step provides the model with the correct answer for each example, enabling it to learn the mapping from image features to the desired output during training. Without labeled data, the model cannot be trained to recognize objects, classify scenes, or detect anomalies in computer vision tasks.

Answer

Adding watermarks to images for copyright protection

Answer

Compressing images to reduce storage costs during training

Answer

Filtering out low-quality or blurry training images

Question 3

A data scientist is building a classification model to predict customer churn. The dataset has only 5% churn cases. The model achieves 95% accuracy on the test set, but upon investigation, the data scientist finds the model predicts 'not churn' for nearly every customer. Which metric should the data scientist primarily use to evaluate the model's performance on this imbalanced dataset?

Accepted Answer

F1 score. In an imbalanced dataset with only 5% churn, a model that predicts 'not churn' for every case achieves 95% accuracy by always guessing the majority class. This accuracy is misleading because it fails to identify any churn cases. The F1 score (option B) is the harmonic mean of precision and recall, making it the primary metric for evaluating classification performance on imbalanced data, as it penalizes both false positives and false negatives and is not skewed by class imbalance.

Answer

Accuracy

Answer

Mean Absolute Error (MAE)

Answer

R-squared

Question 4

A company develops an autonomous vehicle AI system. The system was trained exclusively on data from sunny, dry weather conditions. When the vehicles are deployed in a region that experiences frequent snow and fog, the system fails to correctly identify obstacles, leading to safety risks. Which Microsoft responsible AI principle is most directly violated by this deployment?

Accepted Answer

Reliability and safety. The system fails in snow and fog because it was trained only on sunny, dry data, making it unreliable in those conditions. The Microsoft responsible AI principle of Reliability and safety requires AI systems to perform consistently and safely across their intended deployment environments. Deploying without testing for diverse weather violates this principle by exposing users to safety risks.

Answer

Fairness

Answer

Transparency

Answer

Privacy and security

Question 5

A company plans to use an AI system to analyze employee email communications to identify patterns and improve productivity. The company is concerned about respecting employee boundaries and legal regulations. Which Microsoft responsible AI principle is most important to consider?

Accepted Answer

Privacy and security – protecting employees' personal data and email content.. The scenario involves analyzing employee email communications, which inherently includes sensitive personal data and private correspondence. Microsoft's 'Privacy and security' principle is the most relevant because it mandates that AI systems protect individuals' data and respect boundaries, ensuring compliance with regulations like GDPR and internal privacy policies. Without strong privacy and security safeguards, analyzing email content could violate employee trust and legal requirements, regardless of how fair, reliable, or inclusive the system is.

Answer

Fairness – ensuring the system treats all employees equally.

Answer

Reliability and safety – ensuring the system functions correctly.

Answer

Inclusiveness – ensuring the system works for all employees regardless of communication style.

Question 6

A customer support team wants to automatically analyze incoming emails to (1) determine the overall emotional tone (e.g., frustrated, satisfied) and (2) identify specific key phrases that indicate the reason for contact (e.g., 'return item', 'refund policy'). Which two Azure AI Language features should they use? (Choose two.)

Accepted Answer

Sentiment analysis. Sentiment analysis is the correct choice because it evaluates text to determine the overall emotional tone, such as frustration or satisfaction, by assigning a sentiment score (positive, negative, neutral, or mixed) at the document and sentence level. This directly addresses the requirement to analyze the emotional tone of incoming emails.

Answer

Entity recognition

Answer

Language detection

Question 7

What is the primary use case for Azure AI Document Intelligence's layout model?

Accepted Answer

Extracting the structural layout of documents including tables, text blocks, and positions. Azure AI Document Intelligence's layout model is designed to extract the structural layout of documents, including tables, text blocks, and their spatial positions. This enables downstream processing like OCR, form understanding, and document analysis by preserving the original reading order and layout hierarchy.

Answer

Generating visual layouts for new document templates

Answer

Converting documents between different file formats (PDF to DOCX)

Answer

Checking documents for grammatical and spelling errors

Question 8

What is the maximum output length parameter 'max tokens' used for in Azure OpenAI?

Accepted Answer

The maximum number of tokens in the generated response to control length and cost. The 'max tokens' parameter in Azure OpenAI controls the maximum number of tokens (roughly 0.75 words per token) that the model can generate in a single response. This directly limits the length of the output, which in turn controls both the cost (since Azure OpenAI charges per token) and the response size, preventing excessively long or expensive completions.

Answer

The maximum number of API requests per second

Answer

The maximum number of words in the input prompt

Answer

The maximum number of concurrent users of the model

Question 9

Which Azure AI service is used to index and extract insights from large collections of videos at scale?

Accepted Answer

Azure AI Video Indexer. Azure AI Video Indexer is the correct service because it is specifically designed to ingest large collections of videos, extract metadata (such as transcripts, faces, emotions, and keyframes), and provide searchable insights at scale. Unlike other Azure AI services, Video Indexer combines multiple AI models (speech, vision, and language) into a single pipeline optimized for video content, making it the appropriate choice for indexing and extracting insights from video libraries.

Answer

Azure AI Custom Vision

Answer

Azure Blob Storage media services

Answer

Azure AI Speech transcription only

Question 10

What is the difference between zero-shot, one-shot, and few-shot learning in prompting?

Accepted Answer

Zero-shot uses no examples; few-shot provides multiple examples in the prompt to guide responses. Option B is correct because zero-shot learning involves providing no examples in the prompt, relying solely on the model's pre-trained knowledge to generate a response, while few-shot learning includes multiple examples (typically 2–5) within the prompt to guide the model's output pattern. This distinction is fundamental to prompt engineering in generative AI workloads on Azure, where the number of examples directly influences output consistency and task specificity without retraining the model.

Answer

They refer to how many GPUs are used for model training

Answer

They refer to how many training epochs the model underwent

Answer

Zero-shot is for beginners; few-shot is for experts

Question 11

What is the role of a label (also called target or ground truth) in supervised machine learning?

Accepted Answer

The correct output or answer associated with each training example that the model learns to predict. In supervised machine learning, the label (also called target or ground truth) is the known correct output for each training example. The model uses these labels during training to learn the mapping from input features to outputs, enabling it to make accurate predictions on new, unseen data. This is fundamental to supervised learning, where the algorithm minimizes the error between its predictions and the ground truth labels.

Answer

A category of input features used by the model

Answer

A text description attached to a model explaining what it does

Answer

A tag applied to Azure ML resources for organization

Question 12

A data scientist is training a regression model to predict house prices in Azure Machine Learning. The model uses features like square footage, number of bedrooms, and location (zip code). The data scientist notices that the model has a very low error on the training data but a high error on the test data. Which technique should the data scientist apply during model training to reduce overfitting by penalizing large coefficients?

Accepted Answer

Use a regularization algorithm like Lasso (L1).. Option C is correct because Lasso (L1) regularization adds a penalty equal to the absolute value of the magnitude of coefficients, which can shrink some coefficients to zero, effectively performing feature selection and reducing overfitting. This directly addresses the problem of large coefficients causing the model to fit noise in the training data, leading to high test error.

Answer

Use a smaller test set.

Answer

Apply feature scaling only.

Answer

Increase the number of training epochs.

Question 13

A company develops an AI-powered virtual assistant for customer service. To ensure the assistant can be used by people with visual impairments, the team integrates screen reader compatibility. Which Microsoft responsible AI principle is most directly addressed by this action?

Accepted Answer

Inclusiveness. Option D is correct because integrating screen reader compatibility directly addresses the inclusiveness principle of responsible AI. This principle ensures that AI systems are designed to be accessible and usable by people with diverse abilities, including those with visual impairments, by supporting assistive technologies like screen readers.

Answer

Fairness

Answer

Reliability & Safety

Answer

Privacy & Security

Question 14

A content creator uses Azure OpenAI to generate unique story ideas for a fantasy novel. They want the output to be highly creative and unpredictable, avoiding common clichés. Which parameter should they primarily increase to achieve this?

Accepted Answer

Temperature. Increasing the Temperature parameter makes the model's output more random and less deterministic, which is ideal for generating highly creative and unpredictable story ideas. A higher temperature (e.g., 0.9–1.0) increases the probability of sampling less likely tokens, reducing repetition and clichés.

Answer

Top p

Answer

Frequency penalty

Answer

Presence penalty

Question 15

A construction company uses drone images to survey construction sites. They need an automated system that can identify specific types of heavy equipment (e.g., bulldozers, cranes, excavators) in an image and also draw precise pixel-level outlines around each equipment type. Which Azure Computer Vision capability should they use?

Accepted Answer

Semantic segmentation. Semantic segmentation is the correct capability because it assigns a class label (e.g., bulldozer, crane, excavator) to every pixel in the image, producing precise pixel-level outlines around each equipment type. Object detection only provides bounding boxes, not pixel-level masks, while image classification labels the entire image without localization. OCR is irrelevant as it extracts text, not equipment shapes.

Answer

Object detection

Answer

Image classification

Answer

Optical Character Recognition (OCR)