The label is the value you want the model to predict.
For house-price prediction, the label is the selling price.
For spam detection, the label is spam or not spam.
The target is normally written as y.
In supervised learning your training data contains both the features and the known labels.
The label is the answer you want the model to produce.
Everything else in the row exists to help predict it.
Getting labels is usually the expensive part of a project.
Someone, somewhere, had to decide what the right answer was.
People disagree, get tired, and make mistakes while labelling.
If two labellers often disagree, no model can do better than that confusion.
Wrong labels are worse than missing ones, because the model believes them.
Labels also drift over time as definitions quietly change.
A ticket marked urgent in 2019 may not mean urgent today.
Sometimes the label you want is not the label you have.
You want customer satisfaction, but you only recorded who complained.
That gap is a modelling decision, and it should be a conscious one.
Check how balanced the labels are before you start.
If one answer covers 99 percent of rows, accuracy stops being useful.
Always ask when the label became known, because that is where leakage hides.
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.