Title: Cross-validation for high-dimensional testing in imaging genetics
Authors: Iris Ivy Gauran - King Abdullah University of Science and Technology (Saudi Arabia) [presenting]
Hernando Ombao - King Abdullah University of Science and Technology (KAUST) (Saudi Arabia)
Zhaoxia Yu - University of California Irvine (United States)
Abstract: In estimating the accuracy of a predictive rule, it is essential to understand both the quality of predictions and model selection. Cross-validation is an algorithmic technique extensively used for estimating the prediction error, tuning the regularization parameter, and choosing between competing predictive rules. However, its behavior is non-trivial because of various, complex factors at play. We investigate the performance of cross-validation as a test statistic for high-dimensional testing. We propose a nested cross-validation procedure that performs regularization parameter tuning and accurate variance estimation to formulate the statistic and perform the test rigorously. We present our findings on strategies for improving the statistical power in the high-dimensional as well as the low and equidimensional scenarios. The application of the proposed method to an imaging genetics study and biological data is also presented.