C.Lee Giles – Noisy Time Series Prediction Using a Recurrent Neural Network & Grammatical Inference

Financial forecasting is an example of a signal processing problem that is difficult due to small sample sizes, high noise, and non-linearity. Neural networks can be used in a number of signal processing applications. Neural networks can be used for the processing of high noise, small sample size signals. The new intelligent signal processing method addresses the difficulties. The method proposed uses conversion into a symbolic representation with a self-organizing map. We apply the method to the prediction of daily foreign exchange rates, addressing difficulties with non-stationarity, overfitting, and equal a priori class probabilities, and we find significant predictability in comprehensive experiments covering 5 different foreign exchange rates. The method correctly predicts the direction of change for the next day. Rejecting examples where the system has low confidence in its prediction reduces the error rate to 40%. We show that the symbolic representation can be used to extract symbolic knowledge from neural networks. The operation of the system is explained by these automata. There are automata rules related to well known behavior.