ConceptRolling Origin Evaluation
Rolling-Origin Time Evaluation
A model validation method that respects temporal order and is commonly used in time series forecasting. Its core idea is to select multiple origins from historical data, each time training the model only on data before the origin and predicting data after it, rolling forward step by step to simulate deployment constraints and avoid temporal leakage.
Timeline (last 90 days)
Nov 10
On a public paired cyber-physical dataset, traditional shuffled cross-validation gives a macro F1 score of 0.96, while rolling-origin temporal evaluation gives 0.88 ± 0.06.
Unverified50%