Variable ranking and data reduction in GLM domain

In high and ultra-high dimensional domains, the selection of relevant variables is an important task to reduce the computational cost of the data analysis and to make inference. Here, we present a variable selection algorithm based on the variable ranking which is obtained marginally evaluating the estimated coe!cient associated to each covariate in generalized regression models. The algorithm is applied to gene expression microarray data giving evidence of its performance in ultra-high dimensional context.