What Is Label?
In supervised machine learning, a label is the observed outcome assigned to a training example. In a trading dataset, a label may be a future-return class, the sign of a forward return, whether a price barrier was reached, or the forward return itself. Labels permit supervised models to learn a relationship between features and outcomes.
Labels are often not known until after the prediction time. They may be used to train and evaluate a model once realised, but must never be included as a contemporaneous feature or allowed to overlap improperly with validation data. The terms label and target variable are frequently used interchangeably; label is especially common for categorical outcomes.
See also
Research and literature