View Full Microsoft AI-900 Exam Dumps and Practice Test Dumps.

 

Question 201

A company wants to predict whether a customer will renew a subscription based on previous behavior. The possible results are “renew” and “not renew.” Which machine learning task is appropriate?

1. Regression
2. Classification
3. Clustering
4. Forecasting

Correct Answer: 2

Explanation

Classification is used when a model predicts one category from a set of predefined categories. In this scenario, the possible outcomes are renew and not renew, making the task a binary classification problem. Historical customer information can be used as training features, while the known renewal outcome provides the label. Regression is used for continuous numerical predictions, while clustering identifies groups without predefined labels. Forecasting generally focuses on predicting future values over time. Therefore, classification is the appropriate machine learning task for this subscription-renewal scenario.

Question 202

A machine learning solution groups products based on similarities in their characteristics without being given predefined product categories. Which technique is being used?

1. Classification
2. Regression
3. Clustering
4. Time-series forecasting

Correct Answer: 3

Explanation

Clustering is an unsupervised learning technique that groups similar observations without requiring predefined categories. In this scenario, product characteristics can be analyzed to discover natural groupings, such as products with similar prices, features, or purchasing patterns. Classification would require known categories during training, while regression predicts numerical values. Time-series forecasting focuses on predicting future values based on temporal patterns. Since the product groups are not defined in advance and must be discovered from the data, clustering is the appropriate technique.

Question 203

A model predicts the monthly electricity bill for a household based on its previous usage and other measurements. What type of output does the model produce?

1. A continuous numerical value
2. A category
3. An image
4. A text entity

Correct Answer: 1

Explanation

A predicted electricity bill is a numerical quantity that can take many possible values, making it a continuous numerical output. Machine learning models that predict continuous numerical values commonly use regression techniques. Inputs could include previous electricity consumption, household size, weather conditions, or other relevant features. Categories are used for classification, images represent visual data, and text entities are extracted from language. Since the desired result is an estimated monetary amount rather than a predefined category, the output is a continuous numerical value.

Question 204

A company receives thousands of customer emails and wants to automatically determine the main subjects discussed in them. Which Azure AI Language capability is most suitable?

1. Speech recognition
2. Key phrase extraction
3. Object detection
4. Image classification

Correct Answer: 2

Explanation

Key phrase extraction identifies important words and phrases that represent the main concepts in a text. For customer emails, it can help reveal subjects such as delayed shipments, refunds, product defects, or account problems. This allows organizations to process large amounts of written feedback more efficiently. Speech recognition processes spoken language, while object detection and image classification work with visual information. Because the requirement is to identify the main subjects within written emails, key phrase extraction is the most suitable Azure AI Language capability.

Question 205

A developer wants to create an application that converts written instructions into spoken audio for users. Which capability is required?

1. Speech-to-text
2. Text-to-speech
3. OCR
4. Language detection

Correct Answer: 2

Explanation

Text-to-speech converts written text into spoken audio. It is useful for applications such as accessibility tools, virtual assistants, automated announcements, and navigation systems. The input is written content, and the output is synthesized speech. Speech-to-text performs the reverse process by converting spoken language into written text. OCR extracts text from images, while language detection determines which language is present in a text input. Since the application begins with written instructions and needs to produce spoken output, text-to-speech is the appropriate capability.

Question 206

An organization needs to identify several people and vehicles in a security-camera image and determine where each appears. Which computer vision capability should be selected?

1. Image classification
2. Object detection
3. OCR
4. Image tagging

Correct Answer: 2

Explanation

Object detection identifies objects in an image and determines their locations. In a security-camera scenario, it can identify multiple objects such as people and vehicles and provide their corresponding locations, commonly through bounding boxes. Image classification generally assigns categories to an image without locating each object. OCR extracts written characters, while image tagging identifies concepts or objects represented by tags without necessarily providing precise locations. Because this requirement includes identifying multiple objects and determining where they appear, object detection is the suitable computer vision capability.

Question 207

A generative AI application must answer questions about an organization’s private procedures. The application searches internal documents and supplies relevant passages to the model before generating an answer. What architecture is being used?

1. Retrieval-augmented generation
2. Image segmentation
3. Regression
4. Speech recognition

Correct Answer: 1

Explanation

Retrieval-augmented generation, or RAG, combines information retrieval with generative AI. The application searches an external or private knowledge source, retrieves relevant information, and provides that information to the language model as context. The model can then use the retrieved content when generating its response. This approach is useful for internal knowledge assistants because the required information may be specific to an organization and may not exist in the model’s original training data. The described document-retrieval workflow is therefore an example of RAG.

Question 208

A classification model identifies 80 out of 100 actual positive cases. Which metric is directly represented by this result when the model’s actual positive cases are known?

1. Precision
2. Accuracy
3. Recall
4. Mean absolute error

Correct Answer: 3

Explanation

Recall measures the proportion of actual positive cases that a classification model successfully identifies. If a model identifies 80 of 100 actual positive cases, its recall is 80 percent, assuming the figures describe the complete set of actual positives. Precision instead measures how many predicted positives are actually positive. Accuracy considers correct predictions across all classes, while mean absolute error is generally used for regression. Recall is especially important in scenarios where failing to identify an actual positive case can have significant consequences, such as detecting fraud or certain safety conditions.

Question 209

A model gives highly varied responses when asked the same question repeatedly. Which setting can generally be adjusted to reduce this variation?

1. OCR confidence
2. Image resolution
3. Temperature
4. Number of database tables

Correct Answer: 3

Explanation

Temperature is a generative AI setting that influences the randomness of generated output. Lowering the temperature generally makes responses more predictable and consistent, while higher values can increase variation and creativity. The appropriate value depends on the application because highly deterministic output may be preferable for some tasks, whereas creative applications may benefit from greater variation. OCR confidence is related to text extraction from images, image resolution concerns visual quality, and database table count does not directly control language-model randomness. Therefore, temperature is the relevant setting.

Question 210

A company wants to extract invoice numbers, dates, and totals from a common invoice format using an existing Azure capability without training its own model. Which option is most appropriate?

1. A prebuilt document model
2. A clustering algorithm
3. A regression model
4. A speech model

Correct Answer: 1

Explanation

A prebuilt document model is designed for common document-processing scenarios and can extract supported fields without requiring an organization to train a custom model from scratch. Invoice processing is a common document scenario supported by Azure AI Document Intelligence. A prebuilt model can recognize relevant invoice fields such as invoice numbers, dates, vendor information, and totals. Custom models are more suitable when specialized documents require organization-specific training. Clustering, regression, and speech models address different AI workloads and are not appropriate for this invoice-extraction requirement.

Question 211

An AI system provides a recommendation, but users need information about how the system arrived at the result. Which responsible AI principle is most relevant?

1. Accountability
2. Transparency
3. Inclusiveness
4. Privacy

Correct Answer: 2

Explanation

Transparency focuses on making the behavior and decision-making processes of an AI system understandable to appropriate users and stakeholders. When users need information about how a recommendation was produced, transparency can help provide insight into the factors, logic, or information involved, depending on the system. Accountability concerns responsibility for AI outcomes, inclusiveness focuses on supporting diverse users, and privacy focuses on protecting personal information. Because the primary requirement is understanding how the recommendation was produced, transparency is the most relevant responsible AI principle.

Question 212

An AI application is intended for users with different abilities and backgrounds. What should the development team prioritize?

1. Inclusiveness
2. Increasing model temperature
3. Reducing database storage
4. Removing evaluation data

Correct Answer: 1

Explanation

Inclusiveness involves designing AI systems that can provide useful and accessible experiences for people with different abilities, backgrounds, and circumstances. Development teams should consider diverse users during requirements gathering, design, testing, and deployment. Accessibility needs, language differences, and other relevant user characteristics may need to be considered depending on the application. Increasing temperature affects generative output variation, while database storage and evaluation-data decisions do not directly address diverse user needs. Therefore, inclusiveness should be prioritized when an AI application is intended to serve a broad range of users.

Question 213

A data scientist discovers that two numerical features use very different scales, such as income and age, and wants to normalize their influence on a model. What process can help?

1. Entity recognition
2. Feature scaling
3. Speech synthesis
4. Image tagging

Correct Answer: 2

Explanation

Feature scaling transforms numerical input variables so that their values are represented on comparable scales. This can be useful for machine learning algorithms that are sensitive to differences in feature magnitude. For example, annual income may contain much larger numerical values than age, and scaling can prevent the larger-valued feature from disproportionately influencing certain algorithms. Entity recognition analyzes text, speech synthesis produces audio, and image tagging describes visual content. Therefore, feature scaling is the preprocessing approach relevant to making numerical features more comparable.

Question 214

A company wants to know whether a deployed model’s performance has declined after customer behavior changed significantly. Which activity is most appropriate?

1. Model and data monitoring
2. Removing the model endpoint
3. Deleting evaluation records
4. Increasing image dimensions

Correct Answer: 1

Explanation

Model and data monitoring can help identify changes in incoming data and model performance after deployment. Customer behavior can change over time, causing production data to differ from the data used during model development. Such changes may result in reduced prediction quality. Monitoring performance metrics and relevant characteristics of incoming data can help teams identify potential degradation and determine whether investigation or retraining is needed. Removing endpoints or deleting evaluation records would make monitoring more difficult, while increasing image dimensions does not address general model performance changes.

Question 215

A team evaluates a regression model by calculating the average absolute difference between predicted and actual values. Which metric are they using?

1. Accuracy
2. Precision
3. Mean absolute error
4. Recall

Correct Answer: 3

Explanation

Mean absolute error, or MAE, calculates the average absolute difference between a model’s predicted values and the corresponding actual values. It is commonly used to evaluate regression models because it expresses prediction error in the same general units as the target variable. Accuracy, precision, and recall are primarily associated with classification evaluation. A lower MAE generally indicates that predictions are closer to actual values, although the acceptable level of error depends on the specific business requirement. Therefore, the described calculation represents mean absolute error.

Question 216

A company wants to determine whether an email belongs to the category “spam” or “legitimate.” Which type of machine learning problem is this?

1. Regression
2. Classification
3. Clustering
4. Forecasting

Correct Answer: 2

Explanation

This is a classification problem because the model must assign each email to one of two predefined categories: spam or legitimate. Historical emails can be labeled with their known category and used to train a supervised classification model. Regression is used for predicting continuous numerical values, while clustering discovers groups without predefined labels. Forecasting generally predicts future values associated with time. Since the output consists of predefined categories, binary classification is the appropriate machine learning approach for this email-filtering scenario.

Question 217

A developer wants a generative AI model to produce a response in a specific format, such as a short summary followed by three bullet points. What should the developer include in the prompt?

1. Clear output instructions
2. Random numerical values
3. Unrelated image metadata
4. No task description

Correct Answer: 1

Explanation

Clear output instructions help a generative AI model understand the desired structure and constraints of its response. A developer can specify that the output should contain a short summary followed by three bullet points, for example. Providing explicit requirements can reduce ambiguity and increase the likelihood that the generated response follows the expected format. Random values and unrelated metadata do not provide useful task guidance, while omitting the task description leaves the model without sufficient direction. Clear instructions are therefore an important part of effective prompt design.

Question 218

A company wants to protect personally identifiable customer information while developing an AI solution. Which responsible AI consideration should be addressed?

1. Privacy
2. Object detection
3. Image classification
4. Temperature

Correct Answer: 1

Explanation

Privacy is concerned with protecting personal information used or processed by an AI system. Personally identifiable information should be handled according to appropriate organizational and legal requirements, with controls designed to reduce unauthorized access, disclosure, or inappropriate use. Object detection and image classification are computer vision capabilities, while temperature controls variation in generative AI output. Since the scenario involves protecting customer information, privacy is the responsible AI consideration that should receive direct attention.

Question 219

A model has very poor performance on both its training data and new data because it is too simple to capture the important patterns. What problem does this describe?

1. Overfitting
2. Data encryption
3. Underfitting
4. Model deployment

Correct Answer: 3

Explanation

Underfitting occurs when a model is too simple or insufficiently trained to capture important patterns in the data. As a result, the model can perform poorly on both training examples and unseen data. Overfitting has the opposite pattern: a model performs very well on training data but poorly on new data. Data encryption protects information, while model deployment makes a trained model available for use. When poor performance occurs on both training and new data because the model cannot represent the underlying patterns, underfitting should be investigated.

Question 220

An organization wants to evaluate a classification model while considering both precision and recall together. Which metric can provide a combined measure?

1. Accuracy
2. F1 score
3. Mean absolute error
4. Regression error

Correct Answer: 2

Explanation

The F1 score combines precision and recall into a single measure by using their harmonic mean. It can be useful when an organization wants to consider both false positives and false negatives rather than relying only on overall accuracy. Accuracy measures the proportion of correct predictions across the evaluated examples, while mean absolute error and other regression errors are used for numerical prediction tasks. Therefore, when a classification model needs a combined assessment of precision and recall, the F1 score is an appropriate metric.