基于SICA罚的变量选择及应用

Variable Selection and Application Based on SICA Penalty

高维数据的变量选择一直是统计学领域的热门研究方向。本文研究SICA罚估计在线性模型变量选择中的应用,结合LLA (Local linear approximation)和坐标下降算法给出一种有效的迭代算法,并提出BIC准则选择正则化参数。实际数据的分析表明,与其他变量选择方法相比较,SICA方法在参数估计精度和变量选择方面具有较好的表现。

Variable selection of high-dimensional data has always been a hot research direction in the field of statistics. In this paper, we study the application of SICA penalty estimation in variable selection of linear model, give an effective iterative algorithm combined with LLA (local linear approximation) and coordinate descent algorithm, and propose BIC criterion to select regularization parameters. The analysis of actual data shows that SICA method has better performance in parameter estimation accuracy and variable selection compared with other variable selection methods.