摘要
关于时间序列模型系数估计和定阶问题,前人研究出了对于模型估计有最小二乘估计,逐步回归方法、定阶方法、遗传算法等。但是,这些算法有很多共同缺点,如当变量集较大时,估计误差较大,计算时间较长,不稳定等。为解决以上这些问题,1996年Tibshirani提出了Lasso方法,将模型参数合理地压缩。本研究将Lasso方法应用到BLUED数据集的六组数据上,通过与最小二乘估计进行比较,说明Lasso方法在选择数据时能够应用较少的时间选择出重要的变量特征,同时在分类精度上还能高于最小二乘方法。由此证明Lasso在时间序列建模问题上是一个简单有效的方法。
A lot of work has been done on coefficient estimation and order determination of time series models,and many excellent methods have been developed.For model estimation,there areleast squares estimation,stepwise regression and other methods;for order determination,there is a method of genetic algorithm.However,these algorithms have many common shortcomings,that is,when the variable set is large,the estimation error is big,the calculation time is long,and the algorithm is unstable.In order to solve these problems,Tibshirani proposed Lasso method in 1996 to reasonably compress model parameters.In this paper,Lasso method is applied to six groups of data in BLUED dataset.The result shows that compared with least squares estimation,Lasso method can select important variable features in less time when selecting data,and has higher classification accuracy than least squares method.Thus,it is proved that Lasso method is simple and effective for time series modeling.
作者
王海宽
刘晓宁
WANG Haikuan;LIU Xiaoning(Jincheng Institute of Technology,Jincheng,Shanxi 048026,China;Yulin Real Estate Registration Center,Yulin,Shaanxi 719000,China)
出处
《晋城职业技术学院学报》
2023年第3期92-96,共5页
Journal of Jincheng Institute of Technology
