Bayesian Smoothing, Shrinkage and Variable Selection in Hazard Regression
Bayesian Smoothing, Shrinkage and Variable Selection in Hazard Regression
复制标题
风险回归中的贝叶斯平滑、收缩和变量选择
DOI：
10.1007/978-3-642-35494-6_10
复制
发表时间：
2013
影响因子：
2.2
通讯作者：
T. Kneib
中科院分区：
文献类型：
--
作者：
Susanne Konrath;L. Fahrmeir;T. Kneib
这一贡献涉及一个统一的贝叶斯框架，结合联合收割机正则化的高维线性协变量的影响和半参数平滑的非线性函数的影响，广泛的一类风险回归模型。虽然具有条件高斯平滑先验的惩罚样条形成了估计非参数和灵活时变效应的基础，但高维协变量向量的正则化基于法线先验的尺度混合，包括贝叶斯脊和套索以及收缩方差的尖峰和厚板先验等。这类先验允许我们在模型的预测阶段保持回归系数的条件高斯先验，但为高斯方差引入合适的混合分布以实现正则化。尺度混合属性允许设备通用和自适应马尔可夫链蒙特卡罗模拟算法，用于拟合各种风险回归模型。特别是，基于迭代加权最小二乘建议的统一Metropolis-Hastings算法可以用于正则化和惩罚半参数函数估计。我们通过模拟研究和急性髓性白血病（AML）生存数据的应用程序来展示性能。
This contribution deals with a unified Bayesian framework to combine regularization of high-dimensional linear covariate effects and semiparametric smoothing of nonlinear functional effects for a broad class of hazard regression models. While penalized splines with conditionally Gaussian smoothness priors form the basis for estimating nonparametric and flexible time-varying effects, regularization of high-dimensional covariate vectors is based on scale mixture of normals priors, including among others the Bayesian ridge and lasso as well as a spike and slab prior for shrinkage variances. This class of priors allows us to keep a conditionally Gaussian prior for regression coefficients on the predictor stage of the model but introduces suitable mixture distributions for the Gaussian variance to achieve regularization. The scale mixture property allows to device general and adaptive Markov chain Monte Carlo simulation algorithms for fitting a variety of hazard regression models. In particular, unifying Metropolis-Hastings-algorithms based on iteratively weighted least squares proposals can be employed both for regularization and penalized semiparametric function estimation. We demonstrate performance through simulation studies and an application to data on acute myeloid leukemia (AML) survival.
登录
查看更多内容
影响因子：
20.3
作者：
Metzeler, Klaus H.;Hummel, Manuela;Buske, Christian
通讯作者：
Buske, Christian
影响因子：
1.4
作者：
E. George;R. McCulloch
通讯作者：
E. George;R. McCulloch
DOI：
10.1080/01621459.2012.737742
发表时间：
2012-12-01
影响因子：
3.7
作者：
Scheipl, Fabian;Fahrmeir, Ludwig;Kneib, Thomas
通讯作者：
Kneib, Thomas
DOI：
10.1111/j.1467-9876.2010.00723.x
发表时间：
2011-01-01
影响因子：
1.6
作者：
Kneib, Thomas;Konrath, Susanne;Fahrmeir, Ludwig
通讯作者：
Fahrmeir, Ludwig