Practice for the AI-900: Microsoft Azure AI Fundamentals exam with 14 exam-style practice questions, instant answer reveals, and concise explanations of every correct answer. Topics include: You are training a machine learning model to predict house prices. The dataset includes features like square footage, lo. Follow @CertPunch and visit certpunch.com for more certification practice exams and study content.
Prefer hands-on? Take this round as an interactive practice test — answer every question, get instant feedback, and see your score: Start the AI-900: Microsoft Azure AI Fundamentals practice test →
What you will practice
- You are training a machine learning model to predict house prices. The dataset includes features like square…
- A credit card company wants to minimize fraudulent transactions being approved. Which evaluation metric shoul…
- Which statement best distinguishes a machine learning algorithm from a trained model?
- Which scenario BEST demonstrates the use of Azure's Optical Character Recognition (OCR) capabilities?
- A hospital wants to implement AI for analyzing X-ray images. They need high accuracy but lack resources to tr…
- A healthcare organization needs to process patient records, extract medical information, and make it searchab…
Answers and explanations
Tap a question to expand the answer and the exam reasoning. Try to commit to your own pick first.
Q1. You are training a machine learning model to predict house prices. The dataset includes features like square footage, location, and number of bedrooms. Which method should you use to split the data for training and evaluation?
Answer: C. Randomly split 70% of rows for training and 30% for evaluation.
Randomly splitting data reserves a representative portion, typically around thirty percent, strictly for evaluating model performance on unseen data. Splitting columns is fundamentally incorrect because models require both features and known labels during the training phase to learn patterns.
Q2. A credit card company wants to minimize fraudulent transactions being approved. Which evaluation metric should they prioritize?
Answer: B. Recall
Prioritizing recall minimizes false negatives, effectively ensuring the system catches as many actual fraudulent transactions as possible. While precision reduces false alarms, the business priority in fraud detection is preventing bad actors from successfully completing their unauthorized transactions.
Q3. Which statement best distinguishes a machine learning algorithm from a trained model?
Answer: D. The algorithm is a fixed recipe for processing data, while the trained model is the learned outcome applied for predictions.
The correct option works because an algorithm acts as the fixed mathematical procedure used to process data, while the trained model is the final output used to make predictions. Think of the algorithm as the recipe and the trained model as the baked cake.
Q4. Which scenario BEST demonstrates the use of Azure's Optical Character Recognition (OCR) capabilities?
Answer: A. Extracting handwritten notes from a scanned medical form
Extracting text from scanned medical forms is the correct choice because optical character recognition systems are designed specifically to extract printed and handwritten text from images. Face detection or counting retail products would instead require object detection and spatial analysis features.
Q5. A hospital wants to implement AI for analyzing X-ray images. They need high accuracy but lack resources to train models from scratch. Which Azure service combination would be MOST effective?
Answer: D. Azure AI Vision prebuilt models + Vision Studio
Azure AI Vision prebuilt models provide robust image analysis without requiring custom training, making them ideal for resource-limited environments. Vision Studio complements this by offering a simple interface to test and deploy the models. Building custom neural networks requires too much overhead.
Q6. A healthcare organization needs to process patient records, extract medical information, and make it searchable while maintaining compliance. Which sequence represents the correct Knowledge Mining workflow?
Answer: C. Ingest data → Add AI skills → Create searchable indexes
The correct workflow begins with ingesting raw documents, then applies cognitive skills to extract and enrich the data, and finally creates a searchable index. Applying AI skills before ingesting the data is impossible because the enrichment pipeline needs content to process.
Q7. An AI system for medical diagnosis needs to balance multiple competing requirements. Which statement BEST describes the appropriate approach to model selection?
Answer: C. Balance interpretability and performance based on regulatory requirements and clinical needs
Balancing interpretability and performance ensures the model meets both regulatory standards and clinical accuracy needs. Prioritizing transparency over all else might result in a model that lacks the necessary predictive power for complex diagnoses.
Q8. Which statement BEST describes the relationship between AI, ML, and deep learning?
Answer: B. AI encompasses both rule-based systems and ML approaches
Artificial intelligence is the overarching field that includes both traditional rule-based systems and machine learning approaches. Be careful with absolute distractors; while deep learning is prevalent, computer vision can also utilize classical techniques.
Q9. High ____ occurs when a model performs well on training data but poorly on new data due to memorizing noise.
Answer: C. Variance
High variance indicates overfitting, where a model memorizes training noise instead of learning generalizable patterns. Fairness relates to equitable outcomes, and transparency ensures users understand the system, but neither describes this specific memorization problem.
Q10. A company observes that its AI model performs exceptionally on historical data but exhibits reduced accuracy on new inputs and raises concerns about fairness. Which combination of strategies would best help address both the model's general…
Answer: B. Implement regularization and early stopping while establishing continuous bias audits and engaging stakeholder feedback.
Regularization and early stopping directly prevent overfitting, while continuous bias audits ensure equitable outcomes. Increasing model complexity would worsen generalization, making it a poor choice for fixing poor accuracy on new data.
Q11. A security company wants to enhance doorbell camera footage to better identify license plates in low light. Which type of kernel would be most effective?
Answer: D. Sharpening kernel [[0,-1,0],[-1,5,-1],[0,-1,0]]
A sharpening kernel enhances contrast along edges, making fine details like text characters more distinct in low-light video footage. A blur kernel would smooth out the image and further obscure the license plate characters, while edge detection discards internal details.
Q12. You are designing a prompt for a text summarization task. The model returns summaries that include irrelevant details despite being instructed to "Summarize the text." Which element of prompt engineering is MOST LIKELY missing?
Answer: A. Output Indicator: Format/length constraints.
An output indicator provides specific format or length constraints, which prevents the inclusion of irrelevant details. While the task instruction is present, it lacks the necessary parameters to guide the model effectively.
Q13. A healthcare company uses an AI model to diagnose diseases from X-ray images. They discover the model performs poorly for patients from underrepresented demographics. What is the MOST likely cause of this issue, and how can it be mitigated?
Answer: B. Mitigation: Collect more diverse data and conduct fairness audits.
Poor performance for underrepresented demographics typically stems from unrepresentative training data causing algorithmic bias. Mitigation requires collecting more diverse data and performing regular fairness audits to balance model predictions across all groups.
Q14. Which option does NOT represent a supported method for accessing Azure's OCR capabilities via the Read API?
Answer: A. Querying the OCR service directly through custom SQL commands.
The Azure Read API does not support direct interaction through custom SQL commands. You should remember that Azure AI Vision services are accessed via REST endpoints, language SDKs, or visual tools like Vision Studio.
More AI-900: Microsoft Azure AI Fundamentals drills and other practice exams are on @CertPunch. New rounds drop every few days at certpunch.com.