基于模型树的沪深300指数预测  

CSI 300 index forecast based on model tree

作　　者：林天华[1] 祁旭阳 张倩倩 赵霞[2] LIN Tianhua;QI Xuyang;ZHANG Qianqian;ZHAO Xia(College of Information Technology,Hebei University of Economics and Business,Shijiazhuang 050061,China;Department of Economic Management Experiment,Hebei University of Economics and Business,Shijiazhuang 050061,China)

机构地区：[1]河北经贸大学信息技术学院,石家庄050061 [2]河北经贸大学经管实验中心,石家庄050061

出　　处：《智能计算机与应用》2020年第11期121-125,共5页Intelligent Computer and Applications

基　　金：河北省自然科学基金(F2019207061)。

摘　　要：针对当前智能算法对证券数据预测准确度不高,以及基于最小损失函数的模型树(Model Tree based on Least Loss Function,M TLLF)预测模型不适用于证券数据的预测的问题,本文提出基于最大离差分裂算法的模型树预测方法 (M odel Tree based on Deviation M aximization,M TDM)。使用两组包含完整牛熊市的沪深300指数日收盘价数据进行分组实验验证,得到的均方误差MSE(Mean Squared Error)分别为0.000058和0.000140;均方根误差RMSE(Root Mean Squared Error)分别为0.007634和0.011822;平均绝对百分比误差M APE(M ean Absolute Percent Error)分别为0.011857和0.011348的结果。说明了M TDM预测的稳定性较好,且预测准确度较高。并分别与基于长短记忆神经网络(Long Short-Term M emory,LSTM)和粒子群优化算法(Partical Swarm Optimization,PSO)的预测方法进行实验对比,结果表明MTDM算法的预测误差显著低于前两者。To solve the problem that the accuracy in predicting securities data of current intelligent algorithms is low and the Model Tree based on Least Loss Function(MTLLF) prediction model is not suitable for the prediction of securities data,Model Tree based on Deviation Maximization(MTDM) is put forward.Using two sets of CSI 300 index daily closing price data containing complete bull and bear markets for grouping experimental verification,the obtained MSE(Mean Squared Error) are 0.000058 and 0.000140,the RMSE(Root Mean Squared Error) are 0.007634 and 0.011822 and the MAPE(Mean Absolute Percent Error) are 0.011857 and 0.011348.Indicating that the MTDM prediction has good stability and high prediction accuracy.Compared with the prediction methods based on Long Short-Term Memory(LSTM) and Particle Swarm Optimization(PSO),the results show that the prediction error of MTDM algorithm is significantly lower.

关 键 词：机器学习 模型树 分裂算法 沪深300指数预测 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]