贝叶斯变量选择及模型平均的研究

Research on Bayesian Variable Selection and Model Averaging

对多元线性回归问题中的变量选择进行研究,改进现有的贝叶斯自适应抽样(BAS)方法,在实现整体不放回抽样的前提下,局部引进放回抽样的方法,通过数据仿真发现,同样进行贝叶斯模型平均(BMA),改进后的方法预测效果比改进前的BAS预测效果更好。

This paper mainly studies on the variable selection for multiple linear regression model and is to improve the existing Bayesian adaptive sampling method（BAS）. Not sampling without replacement all the time but partially adopting sampling with replacement, and we can find, through data simulation, that the predictive effect of improved method is better than former one if Bayesian model averaging（BMA） is equally adopted.