Splitting time-series data

Review the temporal data boundaries used for time-series evaluation.

Temporal split

Dividing data by time so earlier observations train and later observations evaluate the model.

Future leakage

Information from a later time enters a feature, training set, or evaluation that should precede it.

Evaluation period

The time window whose observations are reserved for measuring model behavior.

Time-series cross-validation

Repeated time-aware training and evaluation windows that preserve chronological order.