Data augmentation using a combination of independent component analysis and non-linear time-series prediction

In this paper we introduce a new method for filling in gaps in a time series belonging to a set of simultaneously recorded, statistically dependent signals. By combining the properties of the independent component analysis (ICA) transform with those of the dynamical-functionaI artificial neural network (D-FANN), we have developed a data augmentation algorithm that effectively exploits both the temporal history and the mutual dependency between the component signals. This is done by performing the predictions in the ICA-domain, where the signals are expected to maximally independent, whereas the prediction errors, which are used to update the model parameters, are calculated in the observation domain. We have shown that this ICA D-FANN data augmentation algorithm is capable of accurately filling in significant gaps in both synthetic and real time series. Our tests show that the new approach outperforms a predictor based on a standard multi-layer perceptron (MLP) network or a predictor based on the finite impulse response (FIR) network, which works separately on the time series components which have missing values.