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

 

Question 21

Which machine learning approach allows an agent to learn by receiving rewards or penalties for its actions?

1. Supervised learning
2. Unsupervised learning
3. Reinforcement learning
4. Regression

Correct Answer: 3

Explanation

Reinforcement learning is a machine learning approach in which an agent learns by interacting with an environment and receiving feedback based on its actions. The feedback can be represented as rewards or penalties, encouraging the agent to discover strategies that maximize the total reward over time. Unlike supervised learning, reinforcement learning does not require a labeled training example for every decision. It is commonly associated with problems involving sequential decisions, where the action selected at one point can affect future outcomes.

Question 22

A healthcare organization wants to predict whether a patient belongs to a high-risk or low-risk category based on historical records. Which machine learning task is most appropriate?

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

Correct Answer: 1

Explanation

Classification is used when a model needs to assign an input to one of several predefined categories. In this scenario, the expected outcomes are high-risk and low-risk, so the model can learn from historical patient records that contain known categories. The model can then classify new patients based on relevant features. Regression would predict a numerical value, clustering would discover groups without predefined categories, and forecasting generally predicts future values over time. Because the requirement involves selecting between defined categories, classification is appropriate.

Question 23

A company has a large collection of customer data without predefined labels and wants to discover naturally occurring groups. Which learning approach should it use?

1. Supervised learning
2. Unsupervised learning
3. Reinforcement learning
4. Speech recognition

Correct Answer: 2

Explanation

Unsupervised learning is used when data does not contain predefined target labels. Instead of learning to predict known outcomes, the system identifies patterns, relationships, or structures within the available data. Clustering is a common unsupervised learning technique that can discover groups of similar customers based on their characteristics or behavior. Supervised learning requires labeled examples, while reinforcement learning learns through interaction and feedback. Speech recognition is an AI capability rather than a general machine learning approach for discovering unlabeled groups.

Question 24

A streaming service wants to divide users into groups based on viewing habits without specifying the groups in advance. Which technique should it consider?

1. Regression
2. Classification
3. Clustering
4. Speech synthesis

Correct Answer: 3

Explanation

Clustering can identify groups of similar records without requiring predefined labels. A streaming service could provide information such as viewing frequency, preferred genres, watch duration, and other behavioral characteristics. A clustering algorithm could then identify customers who demonstrate similar viewing patterns. These groups might help the organization understand audience segments or design personalized experiences. Classification would require predefined categories, regression predicts numerical values, and speech synthesis converts text into spoken audio. Since the desired groups are not known beforehand, clustering is appropriate.

Question 25

In a machine learning dataset, what is a feature?

1. The value a model is expected to predict
2. An input characteristic used by a model to make a prediction
3. The final accuracy score of a model
4. The name of the machine learning algorithm

Correct Answer: 2

Explanation

A feature is an input characteristic or measurable attribute that a machine learning model uses to identify patterns and make predictions. For example, when predicting house prices, features might include the number of bedrooms, house size, location, and property age. The target or label represents the value the model is trying to predict. Features provide information that helps the model learn relationships within the training data. Understanding which information should be represented as features is an important part of preparing data for machine learning.

Question 26

A dataset contains customer age, annual income, and a column indicating whether each customer purchased a product. In a supervised learning problem, what is the purchase indicator most likely to represent?

1. Feature
2. Label
3. Dataset identifier
4. Model parameter

Correct Answer: 2

Explanation

A label represents the known outcome that a supervised machine learning model is trained to predict. In this example, customer age and annual income can be used as input features, while the purchase indicator provides the expected outcome. During training, the model learns relationships between the features and the label. After training, it can use similar features from new customers to predict whether they are likely to purchase. Labels are therefore essential in supervised learning because they provide the known answers used to train the model.

Question 27

Why is a dataset commonly divided into training and test sets when developing a machine learning model?

1. To evaluate how the trained model performs on unseen data
2. To make every record identical
3. To eliminate the need for features
4. To ensure that the model memorizes the training data

Correct Answer: 1

Explanation

Separating data into training and test sets helps determine whether a trained model can generalize to data it has not previously seen. The training set is used by the model to learn patterns, while the test set is reserved for evaluating the resulting model. If the model performs well only on training examples but poorly on test examples, it may have learned patterns that do not generalize effectively. Keeping the test data separate provides a more realistic indication of how the model may perform in actual use.

Question 28

A model achieves 99 percent accuracy on its training data but only 68 percent accuracy on new test data. Which issue should be investigated first?

1. Underfitting
2. Overfitting
3. Translation
4. Data encryption

Correct Answer: 2

Explanation

The large difference between training performance and test performance can indicate overfitting. An overfitted model may learn the training examples too closely, including patterns that do not represent general relationships in the wider data. Consequently, it can perform extremely well on training data but poorly on unseen examples. Developers can investigate factors such as model complexity, training data quality, and regularization techniques when addressing overfitting. Underfitting generally produces poor performance on both training and unseen data rather than excellent training performance.

Question 29

A machine learning model performs poorly on both its training data and test data because it is too simple to capture important patterns. What problem does this describe?

1. Overfitting
2. Data leakage
3. Underfitting
4. Clustering

Correct Answer: 3

Explanation

Underfitting occurs when a model is unable to learn important patterns from the training data, often because the model is too simple, the selected features are inadequate, or the training process is insufficient. An underfitted model generally performs poorly on both training data and unseen data. This differs from overfitting, where training performance can be high while performance on new data is poor. Developers may address underfitting by improving features, using a more suitable model, or allowing the learning process to capture more relevant patterns.

Question 30

Which Azure service provides tools for developing, training, deploying, and managing machine learning models?

1. Azure Machine Learning
2. Azure DNS
3. Azure Blob Storage
4. Azure Virtual Network

Correct Answer: 1

Explanation

Azure Machine Learning provides a platform for developing and managing machine learning solutions. It supports activities such as preparing data, training models, evaluating model performance, deploying models, and managing machine learning resources. This makes it suitable for organizations that need a structured environment for the machine learning lifecycle. Azure DNS handles name resolution, Azure Blob Storage provides object storage, and Azure Virtual Network provides networking capabilities. When the primary requirement is to build and manage machine learning solutions, Azure Machine Learning is the relevant service.

Question 31

A data science team needs cloud-based computing resources to train a machine learning model without maintaining physical servers. What does Azure Machine Learning provide for this purpose?

1. Managed compute resources
2. Physical desktop computers
3. Domain registration
4. Email hosting

Correct Answer: 1

Explanation

Azure Machine Learning can provide managed compute resources that teams can use for machine learning workloads such as model training and inference. Managed compute reduces the need for organizations to maintain their own physical infrastructure and allows computing resources to be provisioned for machine learning tasks in Azure. The exact compute option can depend on the workload, scale, and performance requirements. Domain registration and email hosting are unrelated services, while physical desktop computers are not the managed cloud computing model described in this scenario.

Question 32

A data scientist wants to create a machine learning workflow by connecting data preparation, model training, and evaluation components through a visual interface. Which Azure Machine Learning capability is designed for this approach?

1. Azure AI Speech
2. Azure Machine Learning Designer
3. Azure AI Vision
4. Azure DNS

Correct Answer: 2

Explanation

Azure Machine Learning Designer provides a visual, drag-and-drop environment for creating machine learning workflows. Users can connect components representing tasks such as data preparation, model training, and evaluation without having to implement the entire workflow through code. This can be useful for users who prefer a visual approach to building machine learning pipelines. Azure AI Speech provides speech-related capabilities, Azure AI Vision focuses on visual information, and Azure DNS provides domain name resolution. The Designer is specifically intended for visual machine learning workflow development.

Question 33

A trained machine learning model is deployed so an application can submit new data and receive predictions. What is this process of using the model to produce predictions called?

1. Training
2. Feature engineering
3. Inference
4. Data labeling

Correct Answer: 3

Explanation

Inference is the process of using a trained machine learning model to generate predictions or outputs for new input data. During training, the model learns patterns from historical examples. Once training is complete, the model can be deployed and used to make predictions on new records through an inference process. Feature engineering prepares useful inputs for the model, while data labeling provides known outcomes for training. Therefore, when an application sends new data to an already trained model and receives a prediction, it is performing inference.

Question 34

An organization has finished training a machine learning model and wants an application to use the model in production. What should the organization do next?

1. Delete the model
2. Deploy the model
3. Remove the training data
4. Convert the model into a spreadsheet

Correct Answer: 2

Explanation

After a machine learning model has been trained and evaluated, it generally needs to be deployed before applications can use it to generate predictions in a production environment. Deployment makes the model available through an appropriate inference mechanism, such as an endpoint. Before deployment, organizations should also consider testing, monitoring, security, and operational requirements. Deleting the model or removing its supporting resources would prevent the application from using it. Therefore, deployment is the appropriate next step when a validated model needs to serve production predictions.

Question 35

A development team has several versions of a machine learning model and wants to identify which version is currently being used in production. Which practice is most useful?

1. Model versioning
2. Image captioning
3. Speech translation
4. Sentiment analysis

Correct Answer: 1

Explanation

Model versioning helps teams maintain and distinguish different versions of machine learning models throughout development and deployment. By assigning identifiable versions, a team can track which model is being tested, which version is deployed, and which version produced particular results. This also supports controlled updates and makes it easier to roll back to an earlier version when necessary. Image captioning, speech translation, and sentiment analysis are AI capabilities that do not provide model lifecycle tracking. Version management is therefore important when multiple model iterations are maintained.

Question 36

A company has five years of monthly electricity consumption data and wants to estimate consumption for the next six months. Which machine learning task is most suitable?

1. Classification
2. Object detection
3. Time-series forecasting
4. Clustering

Correct Answer: 3

Explanation

Time-series forecasting is designed to predict future values based on historical observations collected over time. In this scenario, monthly electricity consumption provides a sequence of time-based measurements that can be analyzed to identify patterns and estimate future consumption. Forecasting can consider trends, recurring patterns, and other characteristics of historical data. Classification would predict categories, object detection analyzes objects within images, and clustering discovers groups without predefined labels. Because the organization needs predictions for future months using historical consumption, time-series forecasting is the appropriate task.

Question 37

A website monitoring system needs to identify unusual traffic patterns that differ significantly from normal activity. Which AI capability is appropriate?

1. Anomaly detection
2. Text-to-speech
3. Image classification
4. Language translation

Correct Answer: 1

Explanation

Anomaly detection can identify observations or patterns that differ significantly from what is considered normal. For website monitoring, the system could analyze traffic volume, request patterns, response behavior, or other relevant measurements and flag unusual activity for investigation. This can help organizations detect unexpected events without requiring every possible abnormal pattern to be explicitly defined in advance. Text-to-speech converts written content into spoken audio, image classification categorizes images, and translation converts content between languages. Anomaly detection is therefore the most relevant capability for unusual website traffic patterns.

Question 38

A company trains a customer prediction model using data collected only from one small region, even though the model will be deployed worldwide. What is a major concern with this training data?

1. It may not be representative of the intended population
2. It contains too many labels
3. It automatically guarantees fairness
4. It eliminates the need for testing

Correct Answer: 1

Explanation

Training data should be representative of the population and conditions in which an AI system will be used. If a model is trained only on data from one small region but deployed worldwide, it may not adequately capture differences in customer behavior, language, demographics, or other relevant characteristics. This can reduce model performance for populations that were poorly represented during training. Organizations should consider the intended deployment population when collecting and evaluating data. Representative training data can help improve the reliability and applicability of the resulting model.

Question 39

Which metric is commonly used to measure the proportion of classification predictions that are correct overall?

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

Correct Answer: 3

Explanation

Accuracy measures the proportion of predictions that a classification model gets correct across the evaluated dataset. It is calculated by comparing the number of correct predictions with the total number of predictions. Accuracy can be useful when the classes are reasonably balanced, although it may be less informative for highly imbalanced datasets. Precision measures the correctness of positive predictions, while recall measures how many actual positive cases were identified. Mean absolute error is generally used for evaluating numerical prediction errors in regression problems.

Question 40

A data scientist wants to evaluate a trained model using examples that were not used to teach the model. Which dataset should be used for this final evaluation?

1. Training dataset
2. Test dataset
3. Feature dataset
4. Label dataset

Correct Answer: 2

Explanation

The test dataset is reserved for evaluating a trained machine learning model using examples that were not used during model training. This helps provide an unbiased indication of how well the model can generalize to unseen data. The training dataset is used to teach the model patterns, while a validation dataset may be used during development for model selection or tuning. A feature dataset and label dataset describe types of information rather than serving as the standard name for the final evaluation set. Therefore, the test dataset is appropriate.