AI-900 Exam Dumps - Microsoft Azure AI Fundamentals

According to the Microsoft Azure AI Fundamentals (AI-900) official study materials and the Microsoft Learn module “Identify features of common machine learning types,” the evaluation phase occurs after training and testing a machine learning model. Evaluation metrics are used to measure how well the model performs when applied to data it has not seen before (the validation data).

The machine learning workflow includes the following key steps:

Data Preparation – Importing, cleaning, and transforming data.

Splitting the Data – Dividing it into training and validation (or test) sets.

Model Training – Using the training data to teach the model patterns or relationships.

Model Evaluation – Assessing the trained model using the validation data and evaluation metrics such as accuracy, precision, recall, F1 score, and root mean square error (RMSE).

As stated in the AI-900 content, evaluation metrics are crucial after testing, as they help determine if the model is accurate enough or if it requires retraining with different parameters or algorithms.

A. After you clean the data → incorrect, as metrics cannot be reviewed before training.

B. Before you train a model → incorrect, since the model has not yet learned patterns.

C. Before you choose the type of model → incorrect, as metrics depend on the model’s output.

Therefore, the verified answer is D. After you test a model on the validation data, which is when you review evaluation metrics to determine model performance and readiness for deployment.

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and Microsoft Learn module “Identify Azure services for conversational AI,” the Azure AI Bot Service is specifically designed to create intelligent conversational agents (chatbots) that can interact with users across multiple communication channels, such as web chat, social media, phone calls, Microsoft Teams, and digital assistants.

In this scenario, customers need the ability to query the status of their orders through various interfaces — including voice and text platforms. Azure AI Bot Service enables this by integrating with Azure AI Language (for understanding natural language), Azure Speech (for speech-to-text and text-to-speech capabilities), and Azure Communication Services (for telephony or chat integration).

The bot can interpret user input like “Where is my order?” or “Check my delivery status,” call backend systems (such as an order database or API), and then respond appropriately to the user through the same communication channel.

Let’s analyze the incorrect options:

B. Azure AI Translator Service: Used for real-time text translation between languages; it doesn’t handle conversation logic or database queries.

C. Azure AI Document Intelligence model: Extracts data from structured and semi-structured documents (e.g., invoices, receipts), not user queries.

D. Azure Machine Learning model: Builds and deploys predictive models, but doesn’t provide conversational or multi-channel interaction capabilities.

Thus, for enabling multi-channel conversational experiences where customers can inquire about order statuses using voice, chat, or digital assistants, the most appropriate solution is Azure AI Bot Service, as outlined in Azure’s AI conversational workload documentation.

Azure AI Form Recognizer extracts information from structured and semi-structured documents. A custom model in Form Recognizer allows an organization to train the system on its specific document layouts and data fields.

As per the AI-900 study guide, a key advantage of a custom model is its flexibility. It can be trained with a set of labeled examples (e.g., invoices, purchase orders, receipts) that match the company’s format. Once trained, the model learns where to locate and extract fields such as invoice numbers, dates, or totals—regardless of layout differences between form types.

Option B is correct because a custom model can be trained to recognize a variety of form types, making it adaptable for diverse business processes.

Options A, C, and D are incorrect:

A: Both prebuilt and custom models are cloud-based; on-premises deployment is not an exclusive feature.

C: Custom models are not cheaper; they may involve additional training costs.

D: Custom models do not always guarantee higher accuracy—accuracy depends on the training data quality.

validation.

In the Microsoft Azure AI Fundamentals (AI-900) study materials, a key concept in machine learning model development is splitting data into subsets for training, validation, and testing. A randomly extracted subset of data from a dataset is most commonly used for validation — that is, for evaluating the performance of the model during or after training.

Here’s how this process works:

Training set – This portion of the dataset is used to train the machine learning model. The model learns patterns, relationships, and parameters from this data.

Validation set – This is a randomly selected subset (separate from training data) used to fine-tune model hyperparameters and evaluate how well the model generalizes to unseen data. It helps detect overfitting — when the model performs well on training data but poorly on new data.

Test set – A final, untouched dataset used to measure the model’s real-world performance after all training and tuning are complete.

By reserving a random subset for validation, data scientists ensure that the model’s performance metrics reflect generalization, not memorization of the training data.

Let’s review the incorrect options:

Algorithms – These are the mathematical frameworks or methods used to build models (e.g., decision trees, neural networks). They are not data subsets.

Features – These are input variables (attributes) used by the model, not randomly selected data subsets.

Labels – These are target values or outcomes the model predicts; again, not data subsets.

Therefore, in alignment with Azure AI-900’s machine learning fundamentals, the correct completion is:

✅ “A randomly extracted subset of data from a dataset is commonly used for validation of the model.”

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and Microsoft Learn module “Identify features of Natural Language Processing (NLP) workloads and services,” the Azure Cognitive Service for Language – Question Answering capability is designed to allow applications to respond to user questions using information from a prebuilt or custom knowledge base. It relies on Natural Language Processing (NLP) to match user queries to the most relevant answers but does not directly execute queries against databases or infer user intent.

“You can use Language Service’s question answering to query an Azure SQL database.” → NOThe Question Answering feature does not connect directly to or query structured databases such as Azure SQL. Instead, it retrieves answers from unstructured or semi-structured content (FAQs, manuals, documents). Querying SQL databases would require traditional database access, not a cognitive service.

“You should use Language Service’s question answering when you want a knowledge base to provide the same answer to different users who submit similar questions.” → YESThis statement is correct and aligns exactly with Microsoft’s official documentation. Question Answering enables organizations to create a knowledge base that can automatically answer repeated or similar customer queries using natural language understanding. For instance, two users asking “How do I reset my password?” and “Can you help me change my password?” would receive the same predefined response.

“Language Service’s question answering can determine the intent of a user utterance.” → NODetermining user intent is handled by Language Understanding (LUIS) or Conversational Language Understanding, not by Question Answering. While both belong to the Language Service, Question Answering focuses on retrieving relevant answers, whereas LUIS focuses on intent detection and entity extraction.

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and the Microsoft Learn module “Describe features of common AI workloads”, clustering is a type of unsupervised machine learning used to group data points that share similar characteristics. In unsupervised learning, the data provided to the model does not have predefined labels or outcomes. Instead, the algorithm identifies inherent patterns or groupings within the dataset based on similarities in input features.

In this scenario, the task is to identify groups of people who have similar purchasing habits. There is no predefined label such as “buyer type” or “purchase category.” The goal is to discover hidden patterns—such as grouping customers by spending behavior, preferred products, or frequency of purchases. This is precisely what clustering algorithms are designed to do.

Clustering is commonly used in:

Customer segmentation for marketing analytics.

Market basket analysis to find associations in purchasing patterns.

Recommender systems to identify similar user profiles.

Anomaly detection when outliers deviate from natural clusters.

Typical algorithms for clustering include K-means, Hierarchical clustering, and DBSCAN. These models analyze multidimensional data to form clusters that maximize intra-group similarity and minimize inter-group similarity.

By contrast:

Classification (A) is a supervised learning method that predicts a categorical label (e.g., whether a customer will churn or not). It requires labeled training data.

Regression (B) is used to predict continuous numeric values (e.g., sales revenue, temperature).

Since the question focuses on discovering groups of similar customers without prior labels, the correct type of machine learning is Clustering.

Object detection is similar to tagging, but the API returns the bounding box coordinates for each tag applied. For example, if an image contains a dog, cat and person, the Detect operation will list those objects together with their coordinates in the image.

According to the Microsoft Azure AI Fundamentals (AI-900) official study guide and Microsoft Learn modules on machine learning concepts, ensuring that the accuracy of a predictive model can be proven requires data partitioning—specifically splitting the available data into training and testing datasets. This is a foundational concept in supervised machine learning.

When you split the data, typically about 70–80% of the dataset is used for training the model, while the remaining 20–30% is used for testing (or validation). The reason behind this approach is to ensure that the model’s performance metrics—such as accuracy, precision, recall, and F1-score—are evaluated on data the model has never seen before. This prevents overfitting and allows you to demonstrate that the model generalizes well to new, unseen data.

In the AI-900 Microsoft Learn content under “Describe the machine learning process”, it is explained that after cleaning and transforming the data, the next essential step is data splitting to “evaluate model performance objectively.” By keeping training and testing data separate, you can prove the reliability and accuracy of the model’s predictions, which is particularly crucial in sensitive domains like clinical or healthcare analytics, where decision transparency and validation are vital.

Option A (Train the model by using the clinical data) is incorrect because you should not train and evaluate on the same data—it would lead to biased results.

Option C (Train the model using automated ML) is incorrect because automated ML is a method for training and tuning, but it doesn’t inherently prove accuracy.

Option D (Validate the model by using the clinical data) is also incorrect if you use the same dataset for validation and training—it would not prove true accuracy.

Therefore, per Microsoft’s official AI-900 study content, the verified correct answer is B. Split the clinical data into two datasets.