Regression in Machine Learning, Explained Simply

You know how to teach classification: show the machine labelled examples, let it learn the categories. I covered that in Classification in Machine Learning.

But what if the answer isn’t a category?

What if it’s a number?

That’s regression — and it quietly powers more business decisions than any other kind of machine learning.

What is regression?

Regression is machine learning that predicts a number.

Not “spam or not spam”. Not “cat or dog”. A quantity: a price, a demand, a duration, a risk score.

Classification answers “which one?” Regression answers “how much?”

You already think in regression

When you estimate what a flat costs from its size, location and age — that’s regression in your head.

You’ve seen hundreds of listings (your training data). You’ve learned how much each factor matters (the weights). You produce a number (the prediction).

Machine learning does the same, with a million listings and no fatigue.

The classics, in plain words

Linear regression

Draw the straight line that best fits the data. Simple, fast, and surprisingly hard to beat. Every extra square metre adds roughly the same amount to the price — that’s the “linear” assumption.

Multiple regression

Same idea, many factors at once: size, location, floor, year built. The model learns how much each one contributes.

Beyond straight lines

Real life curves. Tree-based models and neural networks capture patterns like “price per metre falls in the suburbs but spikes in the centre” — at the cost of being harder to explain.

Where regression runs your business

Sales forecasts. Delivery time estimates. Energy demand. Credit risk scores. Insurance pricing. Marketing spend vs. revenue. “How many support tickets will Monday bring?”

If the question ends in a number, someone somewhere is answering it with regression.

What can go wrong (the part leaders should know)

Correlation isn’t causation. Ice cream sales predict drownings — because summer. A model can predict well and still mislead you about why.

Garbage in, garbage out. A price model trained on 2019 data meets 2026 reality badly. Models age; retraining is maintenance, not failure.

The average hides the extremes. Regression optimises for typical cases. Ask how it behaves on the weird ones — that’s where the risk lives.

Keep building the map

Regression and classification are the two workhorses of supervised learning. The other learning styles — unsupervised and semi-supervised — complete the picture, and AI vs ML vs Deep Learning shows where it all fits.

Prefer learning by doing? My free 10-day AI Crash Course takes you from zero to confident, one email a day.