In order to accommodate the problems of conventional shrinkage estimation methods for their difficulties of statisti-cal inference and determining penalized factors simultaneously,this paper proposed the hierarchy model of Gibbs sampler or weight fused elastic net regression and correspondent Bayesian estimator (BWFEN).The methods determined all penalized factors simultaneously by using Expectation-Maximization algorithm and computes estimator’s standard error according to its empirical posterior distribution,as well as conducts variable selection.Simulation experiments show that BWFEN converges quickly and has low relative prediction error and high variable selection accuracy when being applied to sparse predictive model or the model which has significant group effects.The experiments towards the real dataset “Blog Feed Back”also verifies BW-FEN’s superiority to other shrinkage estimation methods.%针对传统收缩估计中难以进行统计推断以及无法同时确定惩罚因子问题，在权融合弹性网回归（WFEN）的基础上，给出其 Gibbs 层次抽样模型并构造相应的贝叶斯估计量（Bayesian WFEN，BWFEN）。该算法根据 Expectation-Maximization 方法同时确定估计中的两个惩罚因子，并基于回归系数的经验后验分布计算估计量标准误差和进行变量选择。模拟实验表明，BWFEN 的迭代过程具有良好的收敛性，在面对稀疏预测模型或者模型中的预测变量存在群组效应时具有较低的相对预测误差和较高的变量选择精度。在博客回复数预测模型的实际应用中，BWFEN 也显著优于其他收缩估计方法。
展开▼
