[1]陈伟斌,林奕真,王宗跃.股票信息挖掘与LSTM预测[J].集美大学学报(自然版),2020,25(5):385-391.
 CHEN Weibin,LIN Yizhen,WANG Zongyue.Stock Information Mining and LSTM Prediction[J].Journal of Jimei University,2020,25(5):385-391.
点击复制

股票信息挖掘与LSTM预测()
分享到:

《集美大学学报(自然版)》[ISSN:1007-7405/CN:35-1186/N]

卷:
第25卷
期数:
2020年第5期
页码:
385-391
栏目:
数理科学与信息工程
出版日期:
2020-09-30

文章信息/Info

Title:
Stock Information Mining and LSTM Prediction
作者:
陈伟斌1林奕真2王宗跃2
(1.集美大学信息化中心,福建 厦门 361021 ;2.集美大学计算机工程学院,福建 厦门 361021 )
Author(s):
CHEN Weibin1LIN Yizhen2WANG Zongyue2
(1.Informatization Center,Jimei University,Xiamen 361021,China;2.School of Computer Engineering,Jimei University,Xiamen 361021,China)
关键词:
股票预测长短时记忆神经网络(LSTM)回归分析
Keywords:
stock predictionlong short-term memory neural network(LSTM)regression analysis
摘要:
由于受到经济环境、政治政策、市场新闻等多种因素的影响,使得预测股票动态变得极具挑战性。研究了5种常用的预测股价变动的预测方法,通过逐步增加模型的输入维度进行预测分析。首先,建立5种优化的预测模型——基于时间序列的自回归平均模型(ARMA)、灰色预测模型(GM(1,1))、BP神经网络模型(BPNN)、基于改进网格寻优算法的支持向量回归(SVR)模型、基于Tensorflow的长短时记忆神经网络模型(LSTM),研究单一维度的模型输入,即,将各股票的收盘价作为这5种模型的输入。通过实验验证,发现基于LSTM的效果明显优于其他传统机器学习算法。然后,增加模型的输入维度进行研究,即,将影响股价的13个指标作为LSTM模型的输入来预测股价,所得的模型在训练集上的均方误差为0.1438。最后,进一步增加模型的输入维度,即,通过新闻数据挖掘提取14个特征,再结合13个股价指标,以这27个维度特征作为LSTM模型的输入来预测股价,所得的模型在训练集上的均方误差为0.1045。通过实验验证得出,所采用的输入27个维度的方法,比输入13个维度在预测问题上表现得更稳健。
Abstract:
Predicting the dynamic stock price is a challenging task under the influence of economic environment,political policies,market news and other factors.This paper focuses on three ways to predict the stock market,and gradually increases the input dimension of the model for analysis.First,It was considered that single dimension input(closing price)and adopt five optimized forecasting models,autoregressive moving average (ARMA) based on time series,grey prediction (GM(1,1)),back-propagation neural network(BPNN),support vector regression (SVR) based on improved grid search optimization algorithm,and long shorttime memory(LSTM) based on Tensorflow.It is found that the Tensorflow based LSTM is significantly better than other traditional machine learning algorithms.Then,the input dimension of the model is added to study,namely,13 indexes affecting stock price are used as input of LSTM model to predict stock price,and the mean square error of the model on the training set is 0.1438.Finally,the input dimension of the model is further increased.14 features extracted by news mining and 13 stock price indexes are combined (27 dimensions) as input of LSTM model to predict stock price,and the mean square error of the model on the training set is 0.1045.It is concluded that the 27-dimensions model is more robust than that of 13 dimensions.
更新日期/Last Update: 2020-11-04