Section: New Results

Random threshold for linear model selection

Participant : Marc Lavielle.

We have in a previous work introduced a random thresholding method to select the significant, or non-null, mean terms from a collection of independent random variables, and applied it to the problem of recovering the significant coefficients in nonordered model selection.

We have improved this method by introducing a simple modification which removes the dependency of the proposed estimator on a window parameter while maintaining its asymptotic properties [4] . A simulation study suggests that both procedures compare favorably to standard thresholding approaches, such as multiple testing or model-based clustering, in terms of the binary classification risk. An application to the problem of activation detection on functional magnetic resonance imaging (fMRI) data was used to illustrate the performance of the proposed method.