A1591
Title: Forecasting curves portions using motif discovery inspired method
Authors: Yijiang Fan - Sant Anna School of Advanced Studies (Italy) [presenting]
Abstract: A nonparametric method for forecasting in functional data analysis is proposed. We address forecasting the last portion of curves, which can be extended to the imputation of missing portions in the curves. The forecast method is based on the notion of functional motifs, which are patterns that recur in multiple portions of a single curve or in multiple curves. Taking the last portion of a curve as a segment of a candidate motif, the other occurrences of the candidate motif are found; the forecast is the forward projection of the recurrent motif. The feasibility of the proposed forecast method is assessed with diagnostic methods that evaluate whether the last portion of a curve is an occurrence of a motif or not. The performance of the method is examined through simulations of multiple scenarios compared with benchmark methods in functional data forecasting. Eventually, the method is applied to real-world climate data.
