On the application of kernel-based independence tests to variable selection problems

Abstract

Kernel-based generalizations of distance covariance are explored and are applied to variable screening procedures. The flexibility of this association measure allows for the inclusion of models with spherical and hyperspherical data, which are common in various applied research fields such as meteorology, geology, biology, and more. The robustness and adaptability of the proposed method are demonstrated through extensive empirical studies. Overall, the findings suggest that kernel-based distance covariance is a powerful tool for variable selection in high-dimensional datasets.