Long Term Time Series Prediction with Multi-Input Multi-Output Local Learning

Existing approaches to long term time series forecasting are based either on iterated one-step-ahead predictors or direct predictors. In both cases the modeling techniques which are used to implement these predictors are multi-input single-output techniques. This paper discusses the limits of single-output approaches when the predictor is expected to return a long series of future values and presents a multi-output approach to long term prediction. The motivation for this work is the fact that, when predicting multiple steps ahead of a time series, it could be interesting to exploit the information that a future series value could have on another future value. We propose a multi-output extension of our previous work on Lazy Learning, called LL-MIMO, and we introduce an averaging strategy of several long term predictors to improve the final accuracy. In order to show the effectiveness of the method, we present the results obtained on the three training time series of the ESTSP’08 competition.