Course Content
Machine Learning Foundations
14 sections · 70 lessons
What are real-world AI use cases for regression vs classification?
What you need to know
The unit-or-dropdown test
Ask, "What does the model output, exactly?"
- "31 minutes", "82 lakh rupees", "4,200 orders" — a number with a unit, so regression.
- "Fraud", "Billing team", "positive sentiment" — a name from a fixed list, so classification.
Common use cases by industry
| Industry | Regression | Classification |
|---|---|---|
| Food delivery | ETA in minutes, orders per hour per zone | Will this order be cancelled? |
| Banking and UPI | Credit limit, expected loss | Fraud or genuine, approve or reject |
| E-commerce | Units sold next week, price | Will this item be returned? Product category |
| Telecom | Data usage next month | Will this customer churn? |
| Healthcare | Length of hospital stay | Disease present or not in a scan |
| Real estate | Flat price | Is this listing a duplicate? |
| AI products | Tokens and cost per request, latency | User intent, safe or unsafe content |
Many real products use both
A ride-hailing app shows one screen, but behind it:
- Regression estimates the fare (rupees) and the pickup time (minutes).
- Classification predicts whether the driver will accept the ride and whether the payment might fail.
In an interview, naming a product that combines both is a strong way to show you understand how ML is used, not just defined.
Forecasting is a special kind of regression
Predicting next week's sales from past weeks is regression, but the rows are ordered in time. You cannot split randomly, and features must use only the past. The data-splitting section of this course explains why.
A real-life example
A quick-commerce grocery app runs these models every morning:
- Regression: units of milk, bread and eggs needed per dark store for the next 24 hours, so staff stock the right amount. Error is measured in units, and over-stocking dairy wastes money.
- Classification: "will this order be late (over 15 minutes)?", so the app can warn the customer in advance.
- Classification: "is this new product listing in the right category?", to fix seller mistakes.
Each model is small. Together they run the business.
Follow-up questions to expect
- "Is a recommendation system regression or classification?" — Often both or neither. It can predict a rating (regression), a click (classification), or learn to rank items directly (ranking).
- "Is object detection regression or classification?" — Both. It classifies what the object is and regresses the box coordinates around it.
- "Is next-word prediction in an LLM regression or classification?" — Classification over the vocabulary: at each step the model gives a probability for every possible token.