Double SOM for Long-term Time Series Prediction
Geoffroy Simon, Amaury Lendasse, Marie Cottrell, Jean‐Claude Fort, Michel Verleysen, WSOM 2003, Workshop on Self-Organizing Maps, Kyushu Institute of Technology · HAL (Le Centre pour la Communication Scientifique Directe) · 2003
Abstract--- Many time series forecasting problems require the estimation of possibly inaccurate, but long-term, trends, rather than accurate short-term prediction. In this paper, a double use of the Self-Organizing Map algorithm makes it possible to build a model for long-term prediction, which is proven to be stable. The method uses the information on the structure of the series when available, by predicting blocs instead of scalar values. It is illustrated on real time series for both scalar and bloc predictions. 1