Robust sure independence screening for nonpolynomial dimensional generalized linear models

We consider the problem of variable screening in ultra-high-dimensional generalized linear models (GLMs) nonpolynomial orders. Since popular SIS approach is extremely unstable presence contamination and noise, we discuss a new robust procedure based on minimum density power divergence estimator (MDPDE) marginal regression coefficients. Our proposed performs well under pure contaminated data scenarios. provide theoretical motivation for use MDPDEs from both population as sample aspects; particular, prove that are uniformly consistent leading to sure property our algorithm. Finally, propose an appropriate MDPDE-based extension conditional GLMs along with derivation its property. methods illustrated through extensive numerical studies interesting real application.