Selected an Stacking ELMs for Time Series Prediction

Abstract

Extreme learning machine (ELM) has several interesting and significant features. In this paper, a novel pruned Stacking ELMs (PS-ELMs) algorithm for time series prediction (TSP) is proposed. It employs ELM as the level-0 algorithm to train several models for Stacking. And our previously proposed reduce-error pruning for TSP (ReTSP)-Trend pruning technique is used to solve the problem that the level-0 learners might make many correlated error predictions. ReTSP-Trend refers to an evaluation measure for reduce-error pruning for TSP (ReTSP), which takes into account the time series trend and the forecasting error direction. What's more, ELM and simple averaging are used to generate the level-1 model. With the development of PS-ELMs, firstly, those essential advantages of ELM will be naturally inherited. Secondly, those specific defects of ELM are ameliorated to some extent, with the help of ensemble pruning paradigm. Thirdly, ensemble pruning is employed to raise the robustness and accuracy of time series forecasting, making up for the shortages of the existing research. Fourthly, our previously proposed pruning measure ReTSP-Trend is employed in PS-ELMs, which indeed guarantees that the remaining predictor which supplements the subensemble the most will be selected. And finally, the development of PS-ELMs will promote our investigation to the popular ensemble technique of Stacked Generalization. The experimental results on four benchmark financial time series datasets verified the validity of the proposed PS-ELMs algorithm.