Free Certified Artificial Intelligence Practitioner® Practice Questions

Sample Practice Questions

Try these 10 sample questions from the Certified Artificial Intelligence Practitioner® practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1A utility has historical sensor readings paired with confirmed equipment-failure outcomes. How does supervised learning use these records to predict failures?

   1. A. It learns relationships between readings and known outcomes.
   2. B. It groups readings without using the recorded outcomes.
   3. C. It learns actions exclusively from rewards during operation.
   4. D. It applies physical equations without fitting recorded data.

   Answer: A

   Explanation: Google explains that supervised models learn from "data with the known correct results." The confirmed outcomes supply supervision for learning the relationship to sensor readings.

   Reference: What is Machine Learning? — Google for Developers — Supervised learning

2. 2An image dataset contains correctly recorded pixel values, but a review finds that many target labels identify the wrong species. Which data-quality problem has been found?

   1. A. Incorrect feature measurements
   2. B. Missing feature observations
   3. C. Incorrect supervisory labels
   4. D. Duplicated feature dimensions

   Answer: C

   Explanation: Google identifies "Bad labels" as a reliability issue and illustrates a picture assigned the wrong species. Here the erroneous information is the target label, rather than the image features.

   Reference: Datasets: Data characteristics — Quality and reliability of data

3. 3Why does a random forest combine trees trained with bootstrap sampling and randomized feature selection?

   1. A. To make all trees share identical prediction errors
   2. B. To eliminate the need for training target labels
   3. C. To reduce variance by combining diverse tree predictions
   4. D. To force every tree to use the same splitting feature

   Answer: C

   Explanation: The ensemble guide states, "Random forests achieve a reduced variance by combining diverse trees". Sampling and feature randomness reduce dependence among tree errors, allowing aggregation to reduce variance.

   Reference: Ensembles: Gradient boosting, random forests, bagging, voting, stacking — 1.11.2.1, Random Forests

4. 4A requesting application sends newly supplied transaction features across a network and needs a newly computed prediction returned immediately. Which serving arrangement directly supports this request-response interaction?

   1. A. A nightly batch job writes predictions into a shared output file.
   2. B. A periodic batch job refreshes a table of precomputed predictions.
   3. C. A model is embedded in a mobile application's local executable.
   4. D. A prediction microservice computes responses through a REST API.

   Answer: D

   Explanation: Google lists "Microservices with a REST API to serve online predictions." This mode supports request-driven predictions rather than producing an offline batch file.

   Reference: MLOps: Continuous delivery and automation pipelines in machine learning — ML pipeline steps — Model serving

5. 5A public works department needs to predict next month's maintenance expenditure in dollars. Which type of prediction directly matches that output?

   1. A. Binary classification
   2. B. Regression
   3. C. Multiclass classification
   4. D. Clustering

   Answer: B

   Explanation: The source states, "A regression model predicts a numeric value." Expenditure in dollars is a numeric target, rather than a category or a discovered group.

   Reference: What is Machine Learning? — Google for Developers — Regression

6. 6A dataset includes automated bot queries. One model will detect spam; another will improve search results for human users. How should bot-query inclusion be assessed?

   1. A. Keep bot queries for both because more rows improve reliability.
   2. B. Delete bot queries for both because machine activity is invalid.
   3. C. Choose the same filtering rule because the source data is shared.
   4. D. Choose filtering separately based on each model's intended goal.

   Answer: D

   Explanation: Google asks whether bot queries should be included and contrasts spam detection with improving results for humans: "if you're trying to improve search results for humans, then no." Data relevance depends on the task, rather than a universal retain-or-delete rule.

   Reference: Datasets: Data characteristics — Quality and reliability of data — Filtering

7. 7How are trees constructed in histogram-based gradient boosting, in contrast to independently built randomized forest trees?

   1. A. Each tree is fitted without using the target values.
   2. B. Each tree is a copy of the first tree's learned splits.
   3. C. Trees are fitted on shuffled targets and then averaged.
   4. D. Trees are added sequentially to correct earlier errors.

   Answer: D

   Explanation: The guide states, "the decision trees are built sequentially, where each tree is trained to correct the errors made by the previous ones." This sequential error correction distinguishes boosting from the forest construction described in the stem.

   Reference: Ensembles: Gradient boosting, random forests, bagging, voting, stacking — 1.11.2.1, Random Forests — Sequential boosting comparison

8. 8A deployment design calls for the trained model to execute directly inside a mobile device's application. Which serving arrangement describes that design?

   1. A. An embedded model deployed to the mobile device
   2. B. A remote REST endpoint hosted in a cloud service
   3. C. A batch prediction job writing a shared output table
   4. D. A training-only notebook running on a development server

   Answer: A

   Explanation: Google lists "An embedded model to an edge or mobile device." This places inference in the device deployment rather than a remote service or offline batch job.

   Reference: MLOps: Continuous delivery and automation pipelines in machine learning — ML pipeline steps — Model serving

9. 9A research team needs to assign each incoming specimen to one of six predefined species. Which machine learning task matches this requirement?

   1. A. Continuous-value regression
   2. B. Unsupervised cluster discovery
   3. C. Multiclass classification
   4. D. Reward-based policy learning

   Answer: C

   Explanation: Google states, "Multiclass classification models output a value from a class that contains more than two values". Assigning one of six known species is a multiclass prediction.

   Reference: What is Machine Learning? — Google for Developers — Classification

10. 10Two ML tasks each have 1,000 training examples, but that quantity is adequate for only one task. Which explanation is consistent with task-dependent data requirements?

    1. A. Equal input-column counts imply equal training-data requirements.
    2. B. Equal target-class counts imply equal training-data requirements.
    3. C. Data sufficiency depends on the problem and learning setup.
    4. D. Equal serialized file sizes imply equal training-data requirements.

    Answer: C

    Explanation: Google states, "Different datasets for different machine learning programs may require wildly different amounts of examples to build a useful model." Dataset sufficiency is task-dependent; no universal guarantee follows from a fixed count.

    Reference: Datasets: Data characteristics — Quantity of data