Online Program

Accuracy Estimation of the High-Dimensional Variable Selection Methods (303421)

Instability measures have been considered in the statistical literature for evaluating model selection uncertainty. However, low instability measures do not necessarily indicate that the selected model is always trustworthy, since that low instability can also arise when a certain method always tends to select an overly-small model. To overcome the limitation of the instability measures, an estimation method based on F and G measures is proposed in this paper to evaluate the accuracy of the given variable selection methods. The method takes both estimated precision and recall of the variable selection results into account. Large sample properties of the proposed methods are studied. Extensive simulations are conducted to show its very competitive finite sample performance. We further demonstrate the application of our method by using several microarray gene expression data.