基于LSTM的股票价格的多分类预测

Multi-Category Prediction of Stock Price Based on LSTM

为对股票价格的涨跌幅度进行预测,提出一种基于长短期记忆网络(LSTM)的方法。根据股票涨跌幅问题,通过对涨跌幅度做多值量化分类,将其转化成一个多分类问题。将股票的基本交易信息作为特征输入,利用神经网络对其训练,最后对股票的涨跌幅度做分类预测。数据集分沪深300成分股整体类、银行类和证券类三种股票集,实验结果表明该模型在涨跌幅多分类情况下,有比较好的预测效果;同时,在对某一类股票进行预测时,用该类股票的历史交易信息训练的模型要比以整体股票交易信息训练的模型效果好。

In order to predict the rise and fall of stock prices, a method based on long-term and short-term memory network (LSTM) is proposed. According to the stock price increase and decrease, through the quantitative classification of the ups and downs, it is transformed into a multi-classification problem. The basic transaction information of the stock is used as the feature input, and it is trained by the neural network. Finally, the stock’s ups and downs are classified and predicted. The data set is divided into three parts: the whole category set of the Shanghai and Shenzhen 300 constituent stocks, the bank set and the securities. The experimental results show that the model has a better forecasting effect in the case of multi-classification of the ups and downs;at the same time, in a certain kind of stocks when making predictions, the model trained with the historical trading infor-mation of such stocks is better than the model trained with the overall stock trading information.