关键词: 深度自编码器, 受限布尔兹曼机, BP神经网络, 股票预测

Abstract: Stock index data has the characteristics of multiple types, high dimensions, and multiple collinearity. In order to reduce the dimension of the data, eliminate the multiple collinearity, and forecast the stock price. Firstly, the deep autoencoder is constructed  based on the Restricted Boltzmann machine to encode the high-dimensional data to low dimensional space. Then the regression model is established between low-dimensional coding sequence and stock price based on BP neural network. The experimental results show that the ability of the deep autoencoder to extract the feature is better than that of the principal component analysis and factor analysis. Comparing with using original high-dimensional data, the model can reduce the computational cost and achieve higher prediction accuracy by using the coded data.

Key words: deep autoencoder, restricted Boltzmann machine, BP neural network, stock prediction