Online Program

A Sufficient Variable Screening Method with a Nonparametric Dependency Measure for High-Dimensional Data (304990)

Keywords: Variable screening, categorical, nonparametric

A two-stage sufficient variable screening method is proposed to find active predictors, when the number of predictors is much larger than observations. The method utilizes a nonparametric measure that evaluates the dependency between two random vectors. The measure is from a larger class of independence measures using characteristic function. Sample versions and their asymptotic properties using different estimation approaches are developed. This method works especially well for categorical response. Simulation examples and gene expression data analysis illustrate the effective use of our method in detecting active predictors for high dimensional medical data.