Unsupervised learning works on data with no labels attached.
The model tries to discover structure or grouping hiding inside the data.
For example, grouping customers by their buying behaviour.
Those groups are discovered by the algorithm, not handed to it.
Clustering and dimensionality reduction are the two common families.
Unsupervised learning has no answer key.
You hand it data and ask what structure is in there.
Because there is no right answer, there is no simple score.
That makes it much harder to tell whether it worked.
The commonest use is grouping similar things together.
Customers who buy alike, documents on the same topic, users who behave the same way.
The groups it finds are real patterns in the numbers.
Whether they mean anything is a judgement you have to make yourself.
Change the settings and you get different groups, all equally valid mathematically.
The other common use is squashing many columns down into a few.
That helps with speed, with plotting, and sometimes with noise.
Unsupervised results are often a first step rather than the final product.
You find the groups, then a person looks at them and names them.
It also helps when labelling everything would cost too much.
Label a handful of examples from each group instead of all of them.
One question on this card. Every option explains itself once you pick, right or wrong.
55 concepts in this module, then a graded test with every answer explained.