混合多个SVR 模型的金融时间序列预测.doc

文档介绍：收稿日期: 2012-11-29。陈钟国,硕士研究生,主研领域:机器学习。
混合多个SVR模型的金融时间序列预测
陈钟国
(上海交通大学软件学院上海 200240)
基于支持向量回归(SVR)进行金融时间序列预测,使用PSO算法确定SVR超参数,并用实验的方法选择合适的SVR输入向量。为了解决金融时间序列非平稳性导致的单一SVR模型预测精度不稳定的问题,提出一种混合多个SVR模型的预测算法,选取训练数据的不同子集训练出多个SVR模型,采用对多个模型的预测结果加权求和的方法进行预测,各个模型的权重根据其预测误差动态调整。在全球5大股指上的实验表明,该算法的预测能力明显优于单一SVR模型。
支持向量回归金融时间序列预测非平稳性混合多个模型
Financial Time Series Forecasting bining Multiple SVR Models
Chen Zhongguo
(School of Software, Shanghai Jiaotong University, Shanghai 200240, China)
Support Vector Regression (SVR) is employed to forecast financial time series. PSO is used to tune the meta-parameters and kernel parameter of SVR. The input vector and output of SVR is determined empirically. Due to the non-stationary of financial time series, single SVR model suffers from the problem of unstable prediction accuracy. To address this problem, a prediction bining multiple SVR models is proposed. Multiple SVR models are trained using selected subsets of the training data. Prediction is based on the weighted sum of the prediction of these models. Weights of the models are adjusted according to their previous prediction errors. The proposed method is tested using stock price series from five major financial markets. The results show significant enhancement of prediction performance parison with single SVR model.
SVR (Support vector regression) Financial time series forecasting Non-bining mod