Dimensionality Reduction

Accurate stock market prediction can assist in an efficient portfolio and risk management. However, accurately predicting stock price trends still is an elusive goal, not only because the stock market is affected by policies, market environment, and market sentiment, but also because stock price data is inherently complex, noisy, and nonlinear. Recently, the rapid development of deep learning can make the classifiers more robust, which can be used to solve nonlinear problems. This study proposes a hybrid framework using Long Short-Term Memory, Autoencoder, and Deep Neural Networks (LSTM-AE-DNNs). Specifically, LSTM-AE is responsible for extracting relevant features, and in order to predict price movement, the features are fed into two deep learning models based on a recurrent neural network (RNN) and multilayer perceptron (MLP). The dataset used for this is Dow Jones daily stock for 2008-2018, which was used in this article. Besides, to further assess the prediction performance of the proposed model, original stock features are fed to the single RNN and MLP models. The results showed that the proposed model gives the more accurate and best results compared to another. In particular, LSTM-AE+RNN shows a better performance than the LSTM-AE+MLP. In addition, hybrid models show better performance compared to a single DNN fed with the all-stock features directly.