Multi-model integration for long-term time series prediction

Abstract

Long-term (multi-step-ahead) time series prediction is a much more challenging task comparing to the short-term (one-step-ahead) time series prediction. This is due to the increasing uncertainty and the lack of knowledge about the future trend. In this paper, we propose a multi-model integration strategy to 1) generate predicted values using multiple predictive models; and then 2) integrate the predicted values to generate a final predicted value as the output. In the first step, a k-nearest-neighbor (k-NN) based least squares support vector machine (LS-SVM) approach is used for long-term time series prediction. An autoregressive model is then employed in the second step to combine the predicted values from the multiple k-NN based LS-SVM models. The proposed multi-model integration strategy is evaluated using six datasets, and the experimental results demonstrate that the proposed strategy consistently outperforms some existing predictors.