Deep neural networks for stock forecasting

• Main Author: Gao, Kexin
• Other Authors: Lap-Pui Chau
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering::Computer hardware, software and systems; Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/155424
• Institution: Nanyang Technological University

Stock market is an important part of economy. How to effectively predict it to maximize the interests of investors has become a topic of concern to researchers. However, due to the uncertainty and nonstationarity of the stock series, forecasting stock price has become a big challenge. The emergence of neural network enables people to effectively train and predict the long-term correlated nonlinear time series. In this thesis, we use different neural networks to forecast the stock price. We observe the results of the prediction using some conventional neural networks such as Multilayer Perceptron, Recurrent Neural Network, Long Short-Term Memory, Gate Recurrent Unit and Temporal Convolutional Network. These networks all can achieve good performance. But Recurrent Neural Network some time is not stable because of the gradient vanishing phenomenon. Then we use two novel structures, Temporal Attention-Augmented Bilinear Network and a hybrid model named EWT-dpLSTM-PSO-ORELM framework, to do the forecasting. The Hybrid model get a better result than conventional ones. Finally, we propose a new structure named EWT-TCN-PSO-ORELM. The new model’s accuracy is good as the hybrid model but need less time to train and to obtain the prediction results. Keywords: Stock price forcasting, MLP, RNN, LSTM, GRU, TCN, TABL, EWT, PSO, ORELM.