Deep transfer learning based on LSTM model in stock price forecasting
Haoran Xu, Bo Xu, Jie He, Jingrui Bi · 2021
Abstract : Revealing the law of stock price change is a hot topic in financial market research in recent years. Support vector machine, neural network and other machine learning methods are usually used. However, the above methods need a large number of identically distributed training data to ensure the fitting degree of the model. However, when forecasting the stock price with a small amount of historical data, it is often unable to achieve good results. This paper uses a deep transfer learning method based on long-term memory model LSTM. Firstly, a double-layer LSTM network structure is constructed as the basic network training model of deep transfer learning. Then, the part of the network trained in the source domain, is transferred to the target domain by using transfer learning. Finally, the feasibility of this method in stock price prediction is verified by experiments, and the factors influencing the learning effect of the model are analyzed. The experimental results show that the deep transfer learning based on LSTM model has high application value.