ExamPark
Course menuAIF-C01

Exams / AIF-C01 / Study guide / Use cases and ML techniques

Use cases and ML techniques: AIF-C01 study notes and practice questions

What to know about Use cases and ML techniques for the AIF-C01 exam, from the Fundamentals of AI and ML domain (20% of the exam guide), then 3 free questions to test it. Part of the AIF-C01 study guide.

Last verified against current AWS documentation ·How we write our questions

What to know for the exam

Use-case questions describe a business problem and ask which technique fits, or whether AI should be used at all.

Techniques

Technique Output Example
Binary classification One of two categories Will this customer leave: yes or no?
Multi-class classification One of several categories Which department should handle this email?
Regression A number What will this house sell for?
Clustering Groups of similar items, no labels needed Segment customers by buying habits
Forecasting (time series) Future values over time Next month’s demand per store
Anomaly detection Unusual cases Spot fraudulent card payments
Recommendation Items a user may like “Customers also bought”
Computer vision Understanding images and video Detect defects on a production line
Natural language processing Understanding text Find the sentiment of reviews
Generative AI New content Draft product descriptions

When AI and ML help

  • The rules are too complex or unknown to write by hand.
  • There is enough relevant data.
  • Patterns change over time, and the system should adapt.
  • The task is repetitive and at scale, such as reading thousands of documents.

When they don’t

  • Fixed, known rules that must give the exact same result every time (such as a tax calculation): write ordinary code.
  • Not enough data, or no data that represents the problem.
  • A need for certainty where any error is unacceptable and there’s no human review.
  • The cost outweighs the benefit, for example a model for a decision made twice a year.

Common traps

  • Classification vs regression. If the answer is a category, it’s classification; if it’s a quantity, regression.
  • Clustering vs classification. Clustering finds groups without labels; classification predicts known labels.
  • Generative AI for everything. A simple classifier on a table is often cheaper and more accurate than a foundation model for a structured prediction.

Try the 3 practice questions

Practice Use cases and ML techniquesTake the free practice exam

Question 1

Easier ·Link to this question (question 1)

A car leasing company wants to estimate the resale value of each used vehicle when its lease ends, using mileage, age and engine size.

It has five years of past lease returns, each with the price the car eventually sold for.

Which type of ML problem is this?

Choose one answer.

  1. A. Regression, because the model outputs a continuous numeric value, a resale price.
  2. B. Binary classification, because each car either sells above the target or does not.
  3. C. Clustering, because similar vehicles can be grouped into segments.
  4. D. Anomaly detection, because some lease returns sell for unusual prices.
Show the answer and explanation

Answer: A

  • A. Right: Correct. The target is a number on a continuous scale, so this is regression. The training data already holds the final sale prices. Learn more: Linear Learner Algorithm in Amazon SageMaker AI (opens in a new tab)
  • B. Wrong: Binary classification predicts a yes-or-no label. Splitting prices at one target throws away the size of each price, which the company needs.
  • C. Wrong: Clustering groups vehicles without a target value. It does not produce an estimated resale price for a new car.
  • D. Wrong: Anomaly detection flags rare, unusual records. It does not predict the price of a typical car at lease end.

Explanation

When the output is a continuous number, such as a price, the problem is regression. Classification predicts categories, clustering groups data without a target, and anomaly detection looks for unusual records. This question tests matching the type of output to the ML problem.

Study

Question 2

Medium ·Link to this question (question 2)

A payroll team must calculate employee tax withholding each pay run using rates that tax law fixes and publishes every year.

Auditors need each result to be exactly reproducible from the same inputs. A manager suggests training an ML model on last year's payslips instead of writing the calculation.

What is the BEST advice?

Choose one answer.

  1. A. Train a deep learning model, because it will learn the tax rules more precisely than code.
  2. B. Use a foundation model, because it can read the published tax tables and apply them directly to each payslip.
  3. C. Use rule-based code that applies the published rates, since the answer is deterministic and the rules are known.
  4. D. Train a clustering model that groups employees into tax bands, then apply the band rates.
Show the answer and explanation

Answer: C

  • A. Wrong: Deep learning learns approximate patterns from data. It cannot guarantee the exact, auditable result that fixed tax law requires.
  • B. Wrong: A foundation model generates text from learned patterns and can make arithmetic or reading errors. It is not a reproducible calculation engine for legal tax figures.
  • C. Right: Correct. Fixed rules in code give the same answer every time, and auditors can read the logic. An ML model adds nothing when the rules are already known and published.
  • D. Wrong: Clustering finds groups in data without labels and does not calculate tax. Grouping employees does not give the exact withholding amount that payroll must pay.

Explanation

When the rules are fixed, known and must be reproducible, ordinary code is the right tool. Machine learning is useful when the rules are unknown or too complex to write down, not when they are published. This question tests recognizing when ML is the wrong choice.

Study

Question 3

Medium ·Link to this question (question 3)

A telecom operator wants to predict which mobile customers will cancel their contracts next month.

It has five years of structured account data in one table, with a column recording who actually canceled. A developer proposes sending each customer record to a large foundation model with a prompt asking whether the customer will leave.

What is the MOST appropriate approach?

Choose one answer.

  1. A. Send each record to a foundation model, because larger models are more accurate on any table.
  2. B. Train a supervised classification model on the labeled table, then score each customer.
  3. C. Use a diffusion model, because it can generate likely future customer behavior.
  4. D. Use unsupervised clustering, because the cancellation column is not needed at all.
Show the answer and explanation

Answer: B

  • A. Wrong: A foundation model is built for text and open-ended generation. Sending every row through a prompt is slow and costly, and it does not learn from the recorded cancellations.
  • B. Right: Correct. The table already records the outcome for each customer, so a supervised classifier can learn the cancellation pattern and score new customers. It is cheap to run and its accuracy can be measured. Learn more: Built-in SageMaker AI algorithms for tabular data (opens in a new tab)
  • C. Wrong: Diffusion models generate images and similar media from noise. They do not predict whether a specific customer will cancel.
  • D. Wrong: Clustering works without labels, so it discards the cancellation column that makes this a supervised task. It could group customers but would not predict who will cancel.

Explanation

When historical data includes the known outcome, supervised learning is the natural fit, and a classifier on a structured table is efficient. Foundation models and diffusion models solve different problems, and clustering ignores the labels that would make the prediction accurate. This question tests matching the data and outcome to the right ML approach.

Study

Want more on Use cases and ML techniques? The 5 full-length practice sets have 18 more Use cases and ML techniques questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets

Study Use cases and ML techniques

More Fundamentals of AI and ML topics