A supervised-learning example contains input features and a target label or value. Classification predicts a category; regression predicts a numerical quantity. Choose the target to match the decision you need to support.
A feature must be available at prediction time. A ticket’s final resolution code may predict its category extremely well but is unavailable when a new ticket arrives. Including it creates leakage rather than a useful model.
Start with a simple baseline. For classification, a majority-class predictor exposes class imbalance. For regression, predicting a training-set average gives a basic comparison. More complex models should earn their added cost on held-out data.
Exercise: define features and a target for ticket routing and handling-time prediction. Mark any feature that would not exist when the prediction is needed.
Check: you distinguish classification from regression and remove future information from the inputs.