Time-series for ecasting using Bagging techniques and Reservoir Computing

In this paper we present a general procedure to use Bagging techniques for time series processing and forecasting problems. Bagging is one of the most used techniques for combining several predictors in order to produce a highly accurate method. The method uses bootstrap replications of the original training set and for each replicate sample one predictor is generated. After that the method combines the predictors using the majority vote for classification problems and the average function for regression problems. In temporal learning tasks, the order serial of the data precludes to realize bootstrap samples. Here, we present an approach which uses a recurrent neural network to transform the spatio-temporal information of the input data in a new larger space. In this new space is possible to apply bootstrap techniques. In this initial paper, we evaluate our approach on 4 time series benchmarks using linear regressions. Although, the idea presented here is more general and can be used with other kind of statistical methods such that CART, SVM, and so on. The empirical results show the power of this new approach to achieve good performances in temporal learning tasks.