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

 

Question 161

A company wants to determine whether customer comments are positive, negative, or neutral. Which AI workload is most appropriate?

1. Computer vision
2. Speech processing
3. Natural language processing
4. Anomaly detection

Correct Answer: 3

Explanation

Natural language processing, or NLP, enables applications to analyze and work with human language. Customer comments are written language, so NLP capabilities can be used to determine sentiment and extract useful information from the text. Sentiment analysis is one example of an NLP workload. Computer vision processes images and video, speech processing handles spoken audio, and anomaly detection identifies unusual patterns in data. Because the requirement involves understanding written customer comments, natural language processing is the most appropriate AI workload.

Question 162

Which Azure service is designed specifically to provide machine learning capabilities for training, deploying, and managing machine learning models?

1. Azure Machine Learning
2. Azure AI Speech
3. Azure AI Vision
4. Azure AI Language

Correct Answer: 1

Explanation

Azure Machine Learning is a cloud service designed to support the machine learning lifecycle. It provides capabilities for preparing data, training models, evaluating experiments, deploying models, and managing machine learning assets. It can support both code-based and low-code development approaches. Azure AI Speech focuses on speech-related workloads, Vision handles image analysis, and Language provides text and language capabilities. When an organization needs a platform for developing and managing machine learning models rather than a specific pretrained AI service, Azure Machine Learning is appropriate.

Question 163

A model is trained to predict whether a customer will purchase a product. The possible outputs are “yes” and “no.” What type of model output is this?

1. Continuous numerical value
2. Category
3. Image
4. Audio signal

Correct Answer: 2

Explanation

An output such as yes or no represents a category, making it a categorical prediction. In machine learning, predicting among predefined categories is associated with classification. The model could use customer information such as previous purchases, browsing activity, or demographics as input features and produce one of the defined classes as its output. A continuous numerical value would instead be associated with regression. Images and audio signals are different data modalities and do not describe the output type in this scenario.

Question 164

A data scientist has a dataset containing city names such as Lahore, Karachi, and Islamabad. What type of feature are these values?

1. Numerical
2. Time-series
3. Categorical
4. Continuous

Correct Answer: 3

Explanation

Categorical features contain values that represent groups or categories rather than continuous numerical measurements. City names such as Lahore, Karachi, and Islamabad are examples of categorical data because each value identifies a category. Numerical features contain numbers that can represent quantities, while time-series data contains observations associated with points in time. Continuous values can take many numerical values within a range. Correctly identifying categorical features helps determine how data should be prepared before it is provided to a machine learning model.

Question 165

A developer wants to expose a trained machine learning model so that an application can send input data and receive predictions. What is commonly used for this purpose?

1. A model endpoint
2. A training label
3. A feature column
4. A clustering group

Correct Answer: 1

Explanation

A model endpoint provides a way for applications or services to interact with a deployed machine learning model. An application can send input data to the endpoint and receive the model’s prediction as a response, depending on the deployment pattern. Training labels identify expected outcomes during supervised learning, feature columns contain model inputs, and clustering groups represent groups discovered in data. Deploying a model through an appropriate endpoint makes its prediction capabilities accessible to consuming applications.

Question 166

An organization wants to create a model that predicts whether a manufactured component passes or fails inspection using measurements collected from sensors. Which machine learning approach is suitable?

1. Regression
2. Classification
3. Clustering
4. Image generation

Correct Answer: 2

Explanation

Classification is appropriate when the target outcome consists of predefined categories. In this scenario, the component can be assigned to a pass or fail category based on sensor measurements. Historical examples with known inspection results can be used as labeled training data. Regression would be more appropriate if the goal were to predict a continuous measurement, while clustering is used to discover groups without predefined labels. Image generation creates visual content and does not directly address a pass-or-fail prediction task based on sensor measurements.

Question 167

A model identifies 95 percent of all actual fraudulent transactions but also produces some incorrect fraud alerts. Which metric is particularly related to the proportion of actual positive cases that the model successfully identifies?

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 fraud detection model identifies 95 percent of all genuinely fraudulent transactions, its recall is high. Recall is especially important when missing a positive case can have significant consequences. Precision instead measures how many predicted positive cases are actually positive, while accuracy considers all correct predictions. Mean absolute error is generally used for regression. Therefore, recall is the metric most directly associated with identifying a high proportion of actual fraudulent cases.

Question 168

A company wants to build an application that identifies objects in warehouse images and displays their locations using bounding boxes. Which service capability is most relevant?

1. Sentiment analysis
2. Speech synthesis
3. Object detection
4. Language detection

Correct Answer: 3

Explanation

Object detection identifies objects within images and determines where those objects are located. Bounding boxes can be used to represent the detected regions, making object detection suitable for warehouse applications that need to identify and locate products, packages, equipment, or other items. Sentiment analysis and language detection process text, while speech synthesis produces spoken audio from text. Because the requirement involves recognizing physical objects and identifying their positions in images, object detection is the relevant computer vision capability.

Question 169

A company wants to use historical data to predict the number of customer support requests expected next week. Which factor makes this a forecasting problem?

1. The target is associated with a future time period
2. The data must contain images
3. The output must be a category
4. The model cannot use historical observations

Correct Answer: 1

Explanation

Forecasting involves using historical observations and other relevant information to estimate values for a future point or period. In this scenario, previous customer support request volumes can provide patterns that help estimate the number of requests expected next week. The target is therefore associated with a future time period. Forecasting does not require image data or a categorical output, and historical observations are usually an important source of information. The future-oriented nature of the prediction makes this a forecasting scenario.

Question 170

Which activity can help identify whether a machine learning model performs differently for different groups represented in an evaluation dataset?

1. Encrypting the model
2. Comparing performance across relevant groups
3. Increasing storage capacity
4. Converting speech to text

Correct Answer: 2

Explanation

Comparing model performance across relevant groups can help identify potential fairness issues. An organization may examine suitable performance measures for different groups to determine whether the model produces materially different outcomes or error rates. This analysis should use appropriate data and carefully consider the context of the application. Encryption protects data or systems, storage capacity affects infrastructure, and speech-to-text handles spoken language. Evaluating performance across relevant groups is therefore a useful step when investigating whether an AI system behaves fairly.

Question 171

A developer wants to reduce ambiguity when asking a generative AI model to produce a product description in a specific format. Which prompt element is most useful?

1. A clear instruction describing the desired task and output
2. Random unrelated text
3. An empty prompt
4. Removing all output requirements

Correct Answer: 1

Explanation

A clear prompt should communicate what the model needs to accomplish and, when necessary, describe the expected output format. For example, a developer can specify the product information that should be included and request a particular structure such as a short title followed by several descriptive sentences. Clear instructions reduce ambiguity and help the model produce output that is more closely aligned with the application’s requirements. Random text or an empty prompt provides little useful direction, while removing output requirements can increase variation.

Question 172

A generative AI solution needs information that is not included in the model’s original training knowledge. Which approach can provide the model with relevant external information at generation time?

1. Increasing image resolution
2. Changing the file extension
3. Retrieval-augmented generation
4. Removing the user prompt

Correct Answer: 3

Explanation

Retrieval-augmented generation, or RAG, allows an application to retrieve relevant information from an external data source and provide it to the generative model as context. This is useful when the application needs current, private, or domain-specific information that may not be available in the model’s original training knowledge. The retrieved information can help the model generate a response based on the supplied context. Image resolution, file extensions, and removing prompts do not provide the model with relevant external knowledge.

Question 173

A business wants to automatically summarize long customer service conversations into short notes for support agents. Which AI capability is most directly related to this requirement?

1. Object detection
2. Text summarization
3. Image classification
4. Speech synthesis

Correct Answer: 2

Explanation

Text summarization is designed to reduce longer pieces of written content into shorter versions that preserve important information. In a customer-service scenario, summarization can help support agents quickly understand previous conversations without reviewing every message. It can be applied to suitable text content and may be useful for improving workflow efficiency. Object detection and image classification process visual information, while speech synthesis converts written text into spoken audio. Because the requirement is to create concise notes from lengthy written conversations, text summarization is the appropriate capability.

Question 174

A company uses an AI system to make decisions that could affect customers. Which practice helps maintain accountability?

1. Assigning responsibility for the system’s operation and outcomes
2. Removing all records of model decisions
3. Avoiding performance evaluation
4. Allowing the model to operate without oversight

Correct Answer: 1

Explanation

Accountability requires organizations to establish responsibility for how AI systems are developed, deployed, monitored, and used. Assigning appropriate ownership helps ensure that someone is responsible for reviewing system behavior and addressing problems when they occur. Removing records, avoiding evaluation, or operating without oversight can make it more difficult to identify and correct issues. AI systems that influence important decisions should have suitable governance and human oversight appropriate to the context. Establishing clear responsibility is therefore an important part of accountable AI development.

Question 175

A model performs poorly because its training examples do not adequately represent the situations it encounters after deployment. What issue should the team investigate first?

1. Insufficiently representative training data
2. Excessive image resolution
3. Speech synthesis quality
4. User interface color

Correct Answer: 1

Explanation

Training data should be representative of the situations and populations the model is expected to encounter after deployment. If important scenarios are missing or underrepresented, the model may learn patterns that do not generalize effectively to real-world inputs. The team should investigate the composition, quality, coverage, and relevance of the training data. Image resolution may matter for specific computer vision tasks, but it does not explain every type of generalization problem. Speech synthesis and interface color are unrelated to the underlying training-data issue described here.

Question 176

Which capability allows a computer system to convert spoken words into text that can then be processed by an application?

1. Text-to-speech
2. Image tagging
3. Speech-to-text
4. OCR

Correct Answer: 3

Explanation

Speech-to-text converts spoken language into written text. This capability can be used in applications such as voice-controlled assistants, meeting transcription, call-center analysis, and accessibility tools. Once speech has been converted into text, other language-processing capabilities can analyze the resulting content. Text-to-speech performs the reverse operation by generating spoken audio from written text. Image tagging describes visual content, while OCR extracts text from images. Therefore, speech-to-text is the appropriate capability when spoken words need to become processable text.

Question 177

A document-processing application needs to classify different types of company forms before extracting their fields. What can a custom Document Intelligence model help accomplish?

1. Identify and process organization-specific document types
2. Generate spoken audio from the forms
3. Predict stock prices
4. Detect emotions in conversations

Correct Answer: 1

Explanation

Custom Document Intelligence models can be used when an organization has document formats or fields that require specialized processing. A solution can be designed to recognize relevant document structures and extract information according to the organization’s requirements. This is useful for business forms that do not fit common prebuilt document scenarios. Generating speech, predicting stock prices, and detecting emotions involve different AI capabilities. A custom document model is therefore suitable when document processing must be adapted to organization-specific forms and information.

Question 178

A team evaluates a regression model by calculating the average magnitude of its prediction errors without considering whether each error is positive or negative. Which metric can provide this measurement?

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

Correct Answer: 4

Explanation

Mean absolute error calculates the average absolute difference between predicted values and actual values. Because the absolute value is used, positive and negative errors do not cancel each other out. MAE is therefore useful for understanding the typical magnitude of errors made by a regression model. Precision and recall are classification metrics, while accuracy is generally used to describe the proportion of correct classification predictions. For a regression model where the goal is to measure average prediction error in the original value scale, MAE is an appropriate metric.

Question 179

An AI application must read text printed on product labels from photographs. Which capability should be used?

1. OCR
2. Sentiment analysis
3. Speech recognition
4. Clustering

Correct Answer: 1

Explanation

Optical character recognition, or OCR, extracts written or printed characters from images and converts them into machine-readable text. A product-label application can use OCR to read names, codes, prices, or other information visible in photographs. Sentiment analysis is used to understand opinions in text, speech recognition converts spoken language into text, and clustering groups similar data points. Since the input is a photograph containing printed text, OCR is the capability directly suited to the requirement.

Question 180

A company wants to deploy an AI model but first needs to determine whether its predictions meet the required business performance level. What should the team do?

1. Change the model’s name
2. Evaluate the model using suitable metrics and representative data
3. Delete the training records
4. Disable all monitoring

Correct Answer: 2

Explanation

Before deploying an AI model for real-world use, the team should evaluate its predictions using suitable metrics and data that represent the intended workload. The appropriate evaluation method depends on the problem, such as classification or regression, and may include metrics such as accuracy, precision, recall, F1 score, or mean absolute error. Representative evaluation data helps provide a realistic indication of expected performance. Renaming a model, deleting training records, or disabling monitoring does not establish whether the model satisfies the required business performance level.