超高维稀疏二次判别分析两阶段估计方法

Two-stage Estimation Method for Ultra-high Dimensional Sparse Quadratic Discriminant Analysis

文章提出一种两阶段二次判别分析建模方法,该方法将高维协方差阵的估计转化为低维矩阵的估计问题,从而有效解决了超高维二次判别分析计算量大的问题。数值模拟和实际数据分析结果表明,在有限样本情形下,两阶段估计方法在变量选择和分类误差率方面的性能更好。

This paper proposes a modeling method for two-stage quadratic discriminant analysis, which transforms the estimation of high-dimensional covariance matrix into the estimation of low-dimensional matrix, thus effectively solving the problem of large computation of super-high-dimensional quadratic discriminant analysis. The results of numerical simulation and actual data analysis show that the two-stage estimation method has better performance in variable selection and classification error rate under finite samples.