Active subset selection approach to nonlinear modeling of ECG data

In this article we report results concerning an ensembling approach for regression modeling which could be used for data compression and prediction. We train a sequence of models on small subsets of a large data set in order to achieve small computation time and memory consumption. An active learning approach is used to increase the training subset iteratively to cover the full dynamics of the data set without using all observations for the actual training. The algorithm is part of a software toolbox for ensemble regression modeling that is used to demonstrate the performance of this method on examples of measured ECG time series.