基于SVM的多变量股市时间序列预测研究  

ON SVM-BASED MULTI-VARIABLE STOCK MARKET TIME SERIES PREDICTION

作　　者：金桃[1] 岳敏[2] 穆进超[2] 宋伟国[2] 何艳珊[2] 陈毅[2] 

机构地区：[1]吉林广播电视大学教学处理工系,吉林长春130022 [2]兰州大学计算机科学与工程学院,甘肃兰州730000

出　　处：《计算机应用与软件》2010年第6期191-194,209,共5页Computer Applications and Software

摘　　要：目前在股市时间序列预测中,大多数采用单变量时间序列预测算法,导致预测准确度不够高。提出采用基于支持向量机SVM(Support Vector Machines)的多变量股市时间序列预测算法,来提高预测准确度。SVM训练算法中,合适的参数可以使训练模型具有更好泛化能力。交叉验证具有指导参数选择的能力,然而考虑到交叉验证算法效率不高的问题,将其并行化,既达到了参数优选的目的,又避免了传统交叉验证效率低的问题。然后,根据较优参数建立多变量SVM时间序列回归预测模型,进行预测。实验证明,预测平均绝对百分比误差控制在10%以内,并且较之单变量的SVM回归预测有更好的泛化能力。At present,the majority of methods in stock market time series prediction are single-variable time series prediction algorithms,whose prediction accuracies are unsatisfied.In this paper,a multi-variable stock market time series prediction algorithm based on support vector machine（SVM） is raised for improving the accuracy of prediction.Using SVM to train the sample,proper parameters can make the training model have better generalization ability.K-fold cross-validation has the ability to direct the parameter selection.However,considering the inefficiency of K-fold cross-validation,parallel K-fold cross-validation algorithm is proposed.In this way,we achieve the aim of parameter optimization and also avoid low efficiency of traditional K-fold cross-validation.After that,the SVM multi-variable time series regression model is built based on the selected parameters to carry out the prediction.Experimental results show that the prediction of mean absolute percentage error is controlled to less than 10% by this method,and its generalization ability is better than that of support vector machines regression prediction with single-variable.

关 键 词：支持向量机 回归 多变量 交叉验证 并行 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]