A Dataset, Plotted

Here is the whole argument for the Statistical Data domain in one picture. Sixty rows of numbers mean nothing to a human being. The same sixty rows, plotted, fall into groups you can see across a room.

1The dataset

Sixty birds have been measured. For each one we have four numbers: wing length, beak length, body mass and tail length. In the vocabulary of Unit 2, those four are the features, and the kind of bird is the label.

About this data

This is illustrative data, generated once so the picture is the same for everyone — it is an example, not a record of birds anybody caught. The groups in it are real in the sense that matters here: they were checked after generating, and the overlap you will find on tail length is genuinely in the numbers.

2Turning the table into a picture

Give each measurement an axis, and every bird becomes a point. Drag to turn the cloud around and look at it from different sides.

Sixty birds, plotted

Change what each axis measures and drag sideways to turn the cloud. Same sixty rows every time — only the view changes.

tap a name to take it off the chart

Each dot is one bird. Its position is Wing length, Body mass and Beak length — three numbers out of the table, turned into a place in space.

Sixty rows of numbers say nothing to a human being. The same sixty rows plotted in space fall into three obvious groups. Nothing was added — this is the Statistical Data domain doing its one job, which is finding the meaning already sitting in a table.

This is data exploration

What you just did is stage 3 of the AI Project Cycle. Exploration is turning rows of numbers into pictures so that the pattern becomes visible — and the pattern is what tells you which model to build next.

3When one measurement is not enough

Put tail length on an axis and look along it on its own. The sparrows and the starlings overlap — there are starlings with shorter tails than some sparrows. On that measurement alone, no model could reliably tell those two kinds apart.

Now bring wing length in as well. The overlap vanishes, because a starling’s wing is far longer than any sparrow’s. Two features succeed exactly where one failed.

Telling two people apart

“About 170 cm” describes half your class. Add “left-handed” and you have narrowed it to a few. Add “plays the violin” and you have one person. No single fact did the work; the combination did.

Why this matters for modelling

This is why a dataset has many columns, and why data exploration comes before modelling. The features that separate your groups are the ones worth feeding the model — and you find out which those are by looking.

4Take the labels away

Press Hide the labels in the widget above. The colours go, and the three clumps are still plainly there.

The link back to Unit 2

Nobody had to tag those birds for the groups to exist — the grouping is a property of the measurements themselves. Finding groups in data that carries no labels is exactly what a clustering model does, and clustering is unsupervised learning. What you are looking at is the pattern such a model would find.

5From the board papers

Which AI domain would be most suitable for developing a price comparison website?CBSE board exam 2026

1 mark

▶Show the answer

Statistical Data. What goes in is numbers in tables — prices, sellers, dates. Nothing is being read as language and nothing is being seen as an image, so it is neither NLP nor Computer Vision.

A shop has a table of every customer and what they bought over a year, with no categories marked. It wants to find groups of similar customers. Name the AI domain and the type of model, and say why. (Practice)

2 marks

▶Show the answer

Domain: Statistical Data — the input is a table of numbers rather than images or language.
Model: clustering, which is an unsupervised learning model — the data carries no labels and nobody has said what the groups should be, so the model has to discover them from the similarities in the data itself.

6Check yourself

Quick Check

Turning rows of numbers into charts so the pattern becomes visible is which stage of the AI Project Cycle?

Quick Check

Groups are clearly visible in a dataset that carries no labels at all. Which kind of model finds them?

Quick Check

Two kinds of bird overlap on tail length but separate cleanly on wing length. What does that show?

← Previous

📄 CSV and Structured Data

What is Computer Vision? 👁️