2020-10-26t13-39-13z.md

date	title	id
2020-10-26 06:39:13 -0700	Rolling Cross Validation for Time Series problems	2020-10-26t13-39-13z

Due to the sequential nature of the data in time series problems, traditional k-fold cross-validation can be problematic, since adjacent data are not independent as assumed, causing data leakage.

Therefore the typical time series approach when it comes to cross-validation is to apply rolling cross-validation. Here, like in k-fold cross-validation, we split the data in training and validation sets $k$ times. However, the validation sets always follow the training sets chronologically. This way no future observations can be used in constructing the forecast. This is illustrated well in the images below taken from Rob J. Hyndman's website:

• Rob Hyndman's blogpost on the subject
• Stackexchange discussion on the subject