基于RF-MIC-PCA的股票趋势预测
Stock trend prediction based on RF-MIC-PCA
马美琛 1林天华 1赵霞1
作者信息
- 1. 河北经贸大学信息技术学院,河北石家庄 050061
- 折叠
摘要
股票因子具有丰富性、相似性等特点,但对其进行趋势预测较难得到良好效果.针对此问题,提出了一种基于RF-MIC-PCA的股票趋势预测算法.首先,利用随机森林(RF)的基尼指数构建因子与类别间的重要性评分规则,剔除低分因子;然后,利用最大信息系数(MIC)构建因子间的相关性评价方法,并融合主成分分析法(PCA)减少因子冗余度;最后,通过随机森林算法预测的分类准确率作为衡量标准,建立基于RF-MIC-PCA的股票趋势预测算法.为验证算法的有效性,从沪深300中选取10只代表性股票进行实验,结果显示RF-MIC-PCA算法在数据集维度降低了20.45％的同时有效提升了算法的预测性能.另外对沪深300、上证50指数进行趋势预测,准确率分别提高了 4.1％和5.0％,验证了算法的普适性,具有一定的实用价值.
Abstract
Stock factors have characteristics such as richness and similarity,it is more difficult to get good results in trend prediction.To solve the problem,RF-MIC-PCA algorithm is proposed.Firstly,the GI of random forest(RF)is used to construct the importance scoring rules between factors and categories to eliminate low-score factors.Then,the maximal information coefficient(MIC)is used to construct a correlation evaluation method between factors,and integrated the principal components analysis(PCA)to reduce the redundancy of factors.Finally,the stock trend prediction algorithm based on RF-MIC-PCA is established by using the classification accuracy of the random forest prediction as a measure.To verify the effectiveness of the algorithm,10 representative stocks from the CSI 300 are selected for the experiment.The results show that the RF-MIC-PCA algorithm effectively improves the prediction performance of the algorithm while reducing the dimension of the data set by 20.45％.In addition,the trend prediction of the CSI 300 and SSE 50 indices improves the accuracy by 4.1％and 5.0％,which verify the universality of the algorithm and have certain practical value.
关键词
股票趋势预测/随机森林/最大信息系数/主成分分析法Key words
Stock trend prediction/Random forest/Maximal information coefficient/Principal component analysis引用本文复制引用
基金项目
河北省自然科学基金(F2021207005)
出版年
2024
NETL
NSTL
万方数据