Skip to Main Content
Stock price prediction is one of hot areas in neural network application. One critical step in neural network application is network training, that data in business database or data warehouse would be selected and refined to form training data sets. Traditional neural network applications in patterns prediction are mainly based on the assumption that behaviors of participants are consistent through the time of data. However, in real world, entity behaviors may change greatly. As a result, data in training data set may be time-variant, reflecting different behavior patterns in different time. So it is difficult to predict behavior patterns from old data. For example, it is impossible to discover the stock price pattern by the stock exchange data ten years ago, especially in the developing stock market like China. Therefore when data is time variant, data selection will influence the training result Since traditional neural network model lacks in the ability of processing time-variant data, this paper presents an improved neural network model, Amnestic neural network, which simulates human cognitive behavior of forgetting, to solve the problem of cross-temporal data selection. The effective of Amnestic neural network was tested by the application on stock price prediction experiment in the stock market of China.