AI-900: Machine Learning

Topic snapshot

Sample questions

These questions are original IT Mastery practice items aligned to this topic area. They are designed for self-assessment and are not official exam questions.

Question 1

Topic: Describe Fundamental Principles of Machine Learning on Azure

A retailer is building a custom model in Azure Machine Learning to classify product images. The team wants to tune the model before deployment, check performance on images the model did not learn from, and keep a separate test dataset for the final review. What should the team do with the validation dataset?

Options:

• A. Use Azure OpenAI Service to generate labels instead of validating.

• B. Merge it into the training dataset and score the model on that same data.

• C. Use it during tuning to compare model performance on unseen images.

• D. Use Azure AI Language to validate the image categories.

Best answer: C

Explanation: A validation dataset gives the team a way to check model performance on data that was not used to train the model. That makes it the right dataset for comparing tuned versions of the model before deployment, while the separate test dataset stays reserved for the final unbiased check.

Training data is used to teach the model patterns, but measuring accuracy only on that same data can be misleading because the model has already seen those examples. A validation dataset contains separate labeled examples that are not used for learning, so it is used during tuning to compare model versions, adjust settings, and estimate how well the model generalizes to new data. If training results are strong but validation results are weak, the model may be overfitting. After tuning is complete, a separate test dataset is typically used once for the final unbiased evaluation. The key point is that validation helps improve model selection before final testing.

• Combining the validation data with training data removes the independent check and can hide overfitting.
• The option using Azure AI Language is a text-analysis service choice, not a way to evaluate an image classifier.
• The option using Azure OpenAI Service confuses generative AI with model evaluation and does not measure generalization.

Question 2

Topic: Describe Fundamental Principles of Machine Learning on Azure

Which Azure Machine Learning capability automatically tests multiple model approaches, compares their performance, and helps streamline parts of the machine learning workflow?

Options:

• A. Data labeling

• B. Azure Machine Learning designer

• C. Azure AI Vision

• D. Automated machine learning

Best answer: D

Explanation: Automated machine learning in Azure Machine Learning is designed to try different algorithms and training settings, then compare the resulting models. This reduces manual effort in common parts of the model-building workflow.

Automated machine learning is an Azure Machine Learning capability for building predictive models more efficiently. It explores multiple model types and training configurations, evaluates them by a chosen metric, and shows which model performs best. That makes it useful when you want to compare models without manually creating and tuning each experiment.

• It automates repeated training trials.
• It compares model results for you.
• It helps streamline model selection in the ML workflow.

Visual workflow tools, data preparation features, and prebuilt AI services each serve different purposes.

• Visual authoring refers to Azure Machine Learning designer, which helps build pipelines visually but is not the main feature for automatically trying and ranking many models.
• Dataset preparation refers to data labeling, which helps create labeled training data rather than compare trained models.
• Prebuilt vision service refers to Azure AI Vision, which analyzes images and video instead of automating custom ML model experiments.

Question 3

Topic: Describe Fundamental Principles of Machine Learning on Azure

A retail company builds a custom machine learning model to forecast daily product demand from historical sales data. The team needs to manage model versions, deploy the model for predictions, and monitor its performance over time. Which Azure service should they use?

Options:

• A. Azure Machine Learning

• B. Azure AI Vision

• C. Azure OpenAI Service

• D. Azure AI Language

Best answer: A

Explanation: Azure Machine Learning is the Azure service for the lifecycle of custom ML models. It supports managing model versions, deploying models for inference, and monitoring them after deployment, which matches the retail forecasting scenario.

Azure Machine Learning is designed for end-to-end management of custom machine learning solutions on Azure. In this scenario, the company already has a forecasting model and needs high-level lifecycle capabilities: model management, deployment for predictions, and ongoing monitoring. Those are core Azure Machine Learning functions.

Prebuilt Azure AI services are better when you want ready-made AI capabilities such as image analysis, language understanding, or generative text without managing a custom predictive model lifecycle. When the requirement is to work with your own trained model and keep track of it after release, Azure Machine Learning is the appropriate choice.

• Vision mismatch Azure AI Vision is for image-based tasks such as OCR, image analysis, and object detection.
• Language mismatch Azure AI Language focuses on NLP tasks like sentiment analysis and entity extraction, not custom model lifecycle management.
• Generative mismatch Azure OpenAI Service is for generative AI scenarios such as chat, summarization, and content generation.

Question 4

Topic: Describe Fundamental Principles of Machine Learning on Azure

Which business scenario is the best example of using clustering in machine learning?

Options:

• A. Predict next quarter’s sales revenue for each store

• B. Generate product descriptions from catalog keywords

• C. Label insurance claims as fraud or not fraud

• D. Segment customers by similar buying behavior without predefined groups

Best answer: D

Explanation: Clustering is an unsupervised machine learning technique that finds natural groupings in data. Segmenting customers by similar buying behavior fits clustering because the goal is to discover groups based on similarity rather than predict a number, assign a known label, or generate new content.

Clustering is used when you want a model to discover natural groupings in data without labeled outcomes. A classic business example is customer segmentation, where customers are grouped by similar purchasing patterns, preferences, or behaviors. The business can then use those segments for targeted marketing, trend analysis, or service planning.

This differs from classification, which assigns known labels such as fraud or not fraud, and from regression, which predicts a numeric value such as sales revenue. It also differs from generative AI, which creates new content like text or images. When the main goal is to organize similar records into groups that were not defined in advance, clustering is the right technique.

• Predicting sales revenue is regression because the output is a continuous numeric value.
• Labeling claims as fraud or not fraud is classification because the categories are known in advance.
• Generating product descriptions is a generative AI task because it creates new text instead of grouping existing data.

Question 5

Topic: Describe Fundamental Principles of Machine Learning on Azure

A company uses Azure Machine Learning to train a model that predicts home prices. In this context, what are features?

Options:

• A. Expected outputs such as the home price used for training

• B. Measurements such as accuracy or mean absolute error

• C. Groups of similar homes discovered without labeled data

• D. Input variables such as square footage and number of bedrooms

Best answer: D

Explanation: Features are the pieces of input data a model uses to learn patterns and make predictions. In a home-price model, values like square footage and bedroom count are features, while the home price is the label.

In machine learning, features are the input variables provided to the model for each record. The model analyzes these inputs to learn patterns that relate to the value it should predict. In a home-price scenario, columns such as square footage, neighborhood, and number of bedrooms are features because they help estimate the price.

• The value being predicted is the label, not a feature.
• Groups created from unlabeled data are clustering results.
• Accuracy and mean absolute error are evaluation metrics used to assess a model.

If the model reads a value as an input to make a prediction, that value is a feature.

• The option describing the home price used for training refers to the label or target.
• The option describing groups of similar homes refers to clustering output, not model inputs.
• The option describing accuracy or mean absolute error refers to evaluation metrics used after training.

Question 6

Topic: Describe Fundamental Principles of Machine Learning on Azure

A retailer uses Azure Machine Learning to classify photos of returned products as damaged or not damaged. The model scores 98% accuracy on the training images but 60% on a separate validation set. The retailer needs reliable predictions for new photos after deployment. What is the best next step?

Options:

• A. Switch to a chatbot with Azure OpenAI Service.

• B. Retrain with more representative images to reduce overfitting.

• C. Keep the current model because training accuracy is high.

• D. Switch to sentiment analysis with Azure AI Language.

Best answer: B

Explanation: High training accuracy combined with much lower validation accuracy is a classic sign of overfitting. The model learned the training images too closely, so retraining with more representative data is the best response.

Overfitting happens when a model learns details or noise from the training set instead of patterns that generalize. In this scenario, 98% on training images but only 60% on validation images means the model looks strong during training but performs poorly on unseen data, which is exactly the warning sign of overfitting.

• Training accuracy shows how well the model learned the examples it already saw.
• Validation accuracy is a better check for how the model may perform on new photos.
• Retraining with more varied, representative labeled images can help the model generalize better.

Switching to a text or generative AI service would not fix this machine learning problem.

• Sentiment analysis is an NLP workload for text, not image classification.
• A chatbot can generate responses, but it does not correct poor classifier generalization.
• Deploying now ignores the large gap between training and validation results.

Question 7

Topic: Describe Fundamental Principles of Machine Learning on Azure

Which task is an example of generative AI content creation rather than a traditional machine learning pattern-recognition task?

Options:

• A. Predicting next month’s sales from historical data

• B. Classifying customer emails by support category

• C. Identifying whether an image contains a bicycle

• D. Generating a draft product description from a few keywords

Best answer: D

Explanation: Generative AI is used to create new content such as text, images, or code from prompts. The product-description task is content creation, while the other tasks recognize patterns in existing data to classify, predict, or detect.

The key difference is whether the system is creating new content or recognizing patterns in existing data. Generative AI produces original outputs such as summaries, emails, product descriptions, images, or code based on prompts or examples. Traditional machine learning pattern-recognition tasks usually map input data to labels, scores, or predictions.

In this question, generating a draft product description from keywords is generative because the model composes new text. By contrast, categorizing emails is classification, forecasting sales is prediction (often regression), and finding a bicycle in an image is computer vision pattern recognition.

A simple rule: if the system’s main job is to produce new content, think generative AI; if its job is to sort, score, detect, or predict, think traditional ML.

• Email sorting is a classification task because it assigns existing messages to categories.
• Sales forecasting is a predictive ML task because it estimates a numeric future value from past data.
• Image detection is pattern recognition because it identifies visual features already present in the image.

Question 8

Topic: Describe Fundamental Principles of Machine Learning on Azure

A video streaming company is planning several AI initiatives on Azure. Executives want to fund the project that is a machine learning classification scenario. Which proposal best fits?

Options:

• A. Use Azure OpenAI Service to summarize support chat transcripts

• B. Use Azure AI Vision to extract text from cancellation forms

• C. Use Azure Machine Learning to group subscribers by viewing patterns

• D. Use Azure Machine Learning to predict whether subscribers will cancel

Best answer: D

Explanation: Classification predicts categories or labels, such as yes/no outcomes. Predicting whether a subscriber will cancel assigns each customer to one of two classes, so it matches a classification scenario.

A classification model is used when the goal is to predict a discrete label from historical data. In this case, each subscriber is assigned to one of two outcomes: likely to cancel or likely to stay. That makes churn prediction a common binary classification example.

The key idea is:

• Inputs are customer features such as viewing history, billing behavior, or support activity.
• The output is a known category or label.
• The model learns from past examples where the outcome is already known.

By contrast, extracting text is an OCR task, summarizing chats is a generative AI task, and grouping similar customers is clustering. Grouping customers may sound similar, but it finds patterns without predefined labels.

• OCR task The text-extraction proposal is document processing, not label prediction.
• Generative output The summarization proposal creates new text rather than assigning a class.
• Unlabeled grouping The customer-grouping proposal is clustering because it forms segments without known target labels.

Question 9

Topic: Describe Fundamental Principles of Machine Learning on Azure

A shipping company plans to use Azure Machine Learning for a custom solution. It needs to classify photos of damaged packages and learn patterns from short warehouse audio clips. The training data is labeled, and the inputs vary widely in lighting, angle, and background noise. Which technique is the BEST fit?

Options:

• A. Train a deep learning neural network in Azure Machine Learning

• B. Use clustering in Azure Machine Learning

• C. Train a linear regression model in Azure Machine Learning

• D. Use Azure OpenAI Service to generate summaries

Best answer: A

Explanation: Deep learning is the best fit because the company has labeled data and needs to recognize complex patterns in images and audio. Neural-network-based models are commonly used for these data types when simpler models cannot capture the variation.

Deep learning is a machine learning technique based on multi-layer neural networks. It is especially useful for rich data such as images, speech, and text because it can learn complex features automatically from many examples. In this scenario, the company wants a custom model in Azure Machine Learning, has labeled training data, and must handle difficult variation such as lighting changes, camera angles, and background noise. That combination points to deep learning rather than simpler models.

Linear regression is for predicting numeric values, clustering is for finding groups in unlabeled data, and Azure OpenAI Service is mainly for generative tasks such as chat or summarization.

• Regression mismatch linear regression predicts numeric values instead of learning complex image or audio classes.
• Unsupervised mismatch clustering groups similar items without labeled outcomes, which does not match the labeled training data.
• Generative mismatch Azure OpenAI Service is mainly used to create or summarize content, not as the core technique for custom classification.

Question 10

Topic: Describe Fundamental Principles of Machine Learning on Azure

A retail company is using Azure Machine Learning to train a demand forecasting model. The training CSV files already exist in Azure Blob Storage. The team needs a workspace resource that lets training jobs access those files. They do not need to run notebooks, scale training nodes, or deploy predictions yet. Which Azure Machine Learning resource should they use?

Options:

• A. Azure Machine Learning compute cluster

• B. Azure Machine Learning datastore

• C. Azure Machine Learning online endpoint

• D. Azure Machine Learning compute instance

Best answer: B

Explanation: The requirement is to make existing files in Azure storage available to Azure Machine Learning. A datastore is the data-focused resource for connecting the workspace to storage, whereas compute resources run code and endpoints serve predictions.

In Azure Machine Learning, a datastore is used to link the workspace to underlying storage such as Azure Blob Storage. That makes it a data service choice: it provides access to stored training data. Compute resources have a different purpose. A compute instance is for interactive development, and a compute cluster is for scalable training or batch processing. An online endpoint is used after deployment to return real-time predictions. When the main need is to reference existing training files, choose the storage connection resource rather than any compute or inference resource.

• Compute instance is meant for development tasks such as running notebooks, not for representing stored training data.
• Compute cluster provides scalable processing power for training jobs, but it does not serve as the storage connection.
• Online endpoint is used to host a deployed model for real-time inference, which is a later stage than data access.

Continue with full practice

Use the AI-900 Practice Test page for the full IT Mastery route, mixed-topic practice, timed mock exams, explanations, and web/mobile app access.

Try AI-900 on Web View AI-900 Practice Test

Related focused pages

• Free AI-900 Full-Length Practice Exam
• AI Workloads
• Computer Vision
• NLP Workloads
• Generative AI

Free review resource

Read the AI-900 Cheat Sheet on Tech Exam Lexicon, then return to IT Mastery for timed practice.