Supervised vs Unsupervised Learning

Statistical learning, which can be classified into supervised learning and unsupervised learning, refers to a series of techniques for estimating the function f. The function f refers to the function f in $Y=f(X)$.

Input Variables, Output Variables

In $Y=f(X)$,
$X$ is the input variable (Input), and $Y$ is the output variable (Output).
Refer to the diagram below. The box with the question mark is the function f box.

X goes into the function box (in). Therefore, it's input.
Y comes out of the function box (out). Therefore, it's output.
Let's understand it this way.

Supervised vs Unsupervised Learning

Unsupervised Learning has only input variables (Input).
Supervised Learning has both input variables (Input) and output variables (Output).

Let's think simply.
A teacher assigned homework to find the function $f$.
In the case of supervised learning, the teacher will provide both X values and Y values. (In this case, you can infer f.)
However, in the case of unsupervised learning, the teacher will only provide X values. (In this case, you obviously cannot infer f.)

Supervised Learning

Supervised learning with output has a clear purpose.
The goal is to infer f for inference or prediction.

Regression and classification problems, which you've probably heard of often, belong to supervised learning.
Both regression and classification problems aim to predict output variables or explain (infer) the correlation between input and output variables.

Regression problems aim to predict and infer quantitative, continuous output variables.
In contrast, classification problems aim to predict and infer qualitative, categorical output variables.

Unsupervised Learning

What about unsupervised learning without output?
What would happen?
Obviously, it's confusing.

Imagine Go stones spread out in front of you.
White Go stones and black Go stones are mixed and spread across the Go board.

Here, we need to figure something out.
All we're given are the Go stones.
This is what we look like when doing unsupervised learning.
Even in this situation, we find our own things to do.

If white Go stones and black Go stones are spread out, what can we do?
Perhaps we would blankly try to group white stones with white stones and black stones with black stones?
This corresponds to clustering problems in unsupervised learning.

Unsupervised Learning Market Segmentation Example

Market segmentation is a representative example of clustering problems.
Due to privacy protection, sellers cannot match individual customers' gender/interests/age (X) with their spending amount (Y).
This means we have no idea which people purchase our products frequently.
However, we can somewhat infer and group what kind of people visit our smart store.

People visiting the smart store can be grouped into numerous characteristics: 'female' group, BTS-loving group, 20s group, 'Shake Shack' frequent visitors group, iPhone users group, and so on.

If the seller perfectly understood the characteristics of people who buy our products frequently and brought that group to the smart store (highly efficient),
they could achieve tremendous profits (highly effective).
For this purpose, clustering problems and unsupervised learning, which group people's characteristics more precisely and diversely, are also considered important.