Sequential learning of differential trend
Eng Yeow Cheu · 2012
This paper describes a simple learning method to sequentially select recent time series values as features to model the differential trend of a time series. This method is used to solve the First International Competition on Time Series Forecasting (ICTSF) [1]. The objective of ICTSF is to predict eight time series with different time frequency and different forecasting horizon. Experimental result shows viability of the method in multi-step forecasting.