基于优化参数的LS-SVM模型的股票价格时间序列预测

为有效预测股票数据,提高投资者的股市投资能力,降低投资风险,提出一种基于优化机器学习方法的股价时间序列预测方法。对股票序列进行了主成分分析,提取累积贡献率大于95%的主成分作为输入变量,并对比了优化核函数宽度g和正则化参数γ后的LS-SVM和SVM模型的预测效果。运用的支持向量机技术经遗传算法优化参数后,降低了预测的均方误差,提高了预测效果和效率,较其他非线性预测方法,具有泛化能力好、鲁棒性强、预测精度高等优点。最后给出了实证结果分析和研究结论,对有效预测股票数据有一定现实指导意义。

Stock Price Time Series Prediction Based on Optimized Parameters of LS-SVM Model

In order to effectively predict the stock data,improve the investor's stock market investment ability and re-duce the investment risk,a method of stock price time series forecasting based on the optimized machine learning meth-od is proposed. The cumulative contribution rate of the stock sequence is greater than that of the stock 95% of the principal components as input variables,and compared the optimization of the kernel function width Gamma and regular-ization parameters before and after the LS-SVM and SVM prediction effect. The results show that the genetic algo-rithm will greatly improve the prediction effect and efficiency of LS-SVM and SVM. The principal component analysis method can effectively solve the drawbacks of the data dimension. The support vector machine has better generalization ability than other nonlinear prediction methods.,Robustness,simple algorithm and high accuracy of prediction,can ef-fectively predict the stock data. The results show that the combination forecasting method has a great advantage for the prediction of time series.