异质数据下基于变点检验分段偏正态均值回归的参数估计
Parameter Estimation for Segmented Skew-Normal Mean Regression Based on Change Point Detection with Heterogeneous Data
姜喆 1吴艳 2吴刘仓1
作者信息
- 1. 昆明理工大学 理学院,云南 昆明 650500;昆明理工大学 应用统计学研究中心,云南 昆明 650500
- 2. 云南大学 数学与统计学院,云南 昆明 650091
- 折叠
摘要
带有偏斜的异质数据广泛出现在大气科学、生物医学和经济学等领域.目前关于异质偏斜数据建模的方法还很少被提出,且现存的分段模型不能自动的对数据分段,大大限制了分段模型的应用场景.针对异质偏斜数据,提出了一种基于偏正态均值回归的分段模型,且在模型的参数估计部分改进了EM算法M步中的两点梯度下降算法,用显示解替代了文献[24]的迭代算法.使用MIC 信息准则做模型的变点检验,同时估计变点的位置.通过数值模拟表明所提模型和算法的有效性.实例分析表明,所提分段偏正态回归模型的预测精度优于不分段偏正态回归模型下的预测精度,且具有更好的解释性.
Abstract
Heterogeneous data with skew character appear in atmospheric sciences,biomedicine and economics.The existing methods for modeling heterogeneous skewed data are rarely proposed,and the existing segmentation models cannot automatically segment the data,which greatly limits the application scenarios of the segmented models.For heterogeneous skewed data,a segmented model based on skew-normal mean regression is proposed,and the two-point gradient descent algorithm in the M step of the EM algorithm is improved in the parameter es-timation of the model,and the iterative algorithm of the literature[24]is replaced by a numerical solution.The change point detection method of the model selects the MIC information criterion and estimates the position of the change point.Numerical simulations show the effectiveness of the proposed models and algorithms.A real example data analysis shows that the prediction accuracy of the proposed segmented model is better than that of the non-segmented skew-normal model,and has better explanatory performance.
关键词
异质偏斜数据/变点检验/分段偏正态均值回归/EM算法优化/MIC信息准则Key words
heterogeneous skewed data/change-point detection/segmented skew-normal mean regression/EM algorithm optimization/MIC information criterion引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据