Using Deep Unsupervised Method for Stock Prediction

Abstract

AbstractNowadays, with the development of stock market globally, it enables the deployment of the resources in the economy which has played a significant role in the modern society. Due to the demand and supply in the market, the stock prices are driven by a variety of factors. In the financial field such as: stock market prediction, information processing, optimal investment, and transaction strategies, deep learning is also widely used because of its advantages. With that being said, stock market prediction is deemed to be a hot topic in the financial sector. In this study, generative adversarial network (GAN) model-unsupervised deep learning methodology is proposed for stock price prediction. The GAN model consists of two layers, the Long Short-Term Memory (LSTM) which is used for prediction of stock prices (Generator) and Bidirectional Long Short-Term Memory (Bi-LSTM) which is used for the discriminator. LSTM is based on traded stock data and produces fake data like distributed data, while the differentiation layer is designed by Bi-LSTM algorithm to differentiate real and fake stock data. We did the experiment on the AMZN (Amazon) index, and the experimental results show that our proposed model could achieve better results in predicting stock prices in comparison to many other predictive models.KeywordsUnsupervised deep learningStock predictionGANLSTMBi-LSTM