When you are dealing with data with both an input and an output, you are dealing with supervised learning. Think of every course you have ever taken: the teacher teaches you the relevant rules with some examples and then you apply the same rules to new examples, which usually are far more difficult than the teacher's one, but this is all another story...

In a more formal way, we talk about supervised learning when our model relies on a set of x, the previously mentioned explanatory variables, and a corresponding set of y. This translates into datasets showing for each record the set of x and the corresponding y.

Starting from cases where the variable to be explained is binary that can be assumed as either one or another value, it is said that for each record a label is attached to it, showing if it has a good output or not. That is why this kind of data is also called labeled.  

A great example of labeled data is our customer dataset. For each record we have a label, the default_flag column, saying if the given set of explanatory variables resulted in a good result (the customer paying) or a bad result (the customer not paying).