摘要
技术指标广泛应用于股票市场的预测分析,不同特征组合对预测效果产生不同影响。为了提高股票趋势预测的准确度,提出一种两层特征选取及预测方法。第一层特征选取以特征子集区分度衡量准则——DFS为评价标准,第二层特征选取以分类器分类效果为评价准则,两层特征选取均采用二进制粒子群(BPSO)算法对特征空间进行搜索。通过第一层特征选取可以高效剔除部分非预测相关特征,在保留预测特征集信息的基础上缩小特征集规模;第二层特征选取可以准确选择出具有较好预测效果的特征子集。实验数据为2015~2016年上海证券综合指数,结果表明,DFS-BPSO-SVM预测模型相比于其它4种特征选取及预测模型,具有更好的预测效果。
Technical indicator was widely used in stock predicting.Different combination of indicator have an effect on predicting performance.In order to improve stock price trend predicting performance,this study proposes a new predicting model that is Binary Particle Swarm Optimization combined with Support Vector Machine and DFS criterion（DFS-BPSO-SVM）predicting model.It＇s a two step feature selection predicting model.In first step,DFS criterion is used for feature selection and we got suboptimal feature subset.After this process,redundant features have been removed and the scale of the feature set becomes smaller.In second step,BPSO-SVM is used for feature selection on suboptimal feature subset and we got best feature subset which leads to best stock trend predicting performance.Based on best feature subset,sample set is constructed for stock trend predicting.In this study,the target is to predict 2015-2016 Shanghai securities composite index daily movement.The experiment results indicate that DFS-BPSO-SVM predicting model have a better performance on stock price and index daily movement than another 4 predicting model.
出处
《软件导刊》
2017年第12期147-151,共5页
Software Guide
基金
河南省基础与前沿技术研究项目(152300410103)
河南省教育厅科学技术研究重点项目基础研究计划(13A510330)
大学生创新创业训练计划项目(201510460029)
