In research of time series forecasting, a lot of uncertainty isudstill related to the question of which forecasting method to use in whichudsituation. One thing is obvious: There is no single method that performsudbest on all time series. This work examines whether features extractedudfrom time series can be exploited for a better understanding of differentudbehaviour of forecasting algorithms. An extensive pool of automaticallyudcomputable features is identified, which is submitted to feature selectionudalgorithms. Finally, a possible relationship between these features andudthe performance of forecasting and forecast combination methods for theudparticular series is investigated.
展开▼
