The question behind a prediction
Imagine predicting how long a trip will take before it begins. You might know the distance, departure time and weather. You do not yet know the actual duration. Those known values are features; the duration you later observe is the label.
Try changing what the model sees
Our toy model estimates duration as 5 + 3 × distance. Change the distance below, then change the observed duration. The prediction responds to the distance but not the label: the label is used to check the prediction afterward. This fixed formula does not train itself. Follow the boxes from the distance input to the prediction. The observed label is separate and does not feed the rule.
INTERACTIVE EXPLORATION
Change the input, then compare
An illustrative model estimates trip time as 5 + 3 × distance. Change distance or the observed duration. Does changing the label alter the prediction?
Why timing matters
A data table could contain the actual arrival time, but that information is unavailable before departure. Using it to make the prediction would leak information from the future. A good feature must be available at the moment the model is meant to act.
Check your understanding
Is observed trip duration a feature or a label when predicting it before a trip?
It is the label. Using the completed trip duration as an input at prediction time would leak the answer.