Evaluation of the Performance of Time Series Forecasting - Business Intelligence with Data Mining

Time series forecasts are evaluated using statistics similar to those of supervised learning for regression. However, there are two very important differences. First, the benchmark comparison takes into consideration the unique sequence of a time series, so there are various default forecast models, depending on whether there is seasonality in the data. Second, models cannot be validated using split or cross validation, whose randomnization would destroy the sequence of the time series; sliding window validation is thus normally used.

[@WahranRai]

Very well explained: clear and precise !