Recurrent Neural Network with Kernel Feature Extraction for Stock Prices Forecasting

A two-stage neural network architecture constructed by combining recurrent neural network (RNN) with kernel feature extraction is proposed for stock prices forecasting. In the first stage, kernel independent component analysis (KICA) and kernel principal component analysis (KPCA) are used as feature extraction. In the second stage, RNN with kernel feature extraction is used to regression estimation. By examining the stock prices data, it is shown that (1) RNN with feature extraction outperforms single RNN; (2) RNN with kernel performs better than those without kernel