Coordinate ascent for penalized semiparametric regression on high-dimensional panel count data

This paper explores a fast algorithm to select relevant predictors for the response process with panel count data. Based on the lasso penalized pseudo-objective function derived from an estimating equation, the coordinate ascent accelerates the estimation of regression coefficients. The coordinate ascent algorithm is capable of selecting relevant predictors for underdetermined problems where the number of predictors far exceeds the number of cases. It relies on a tuning constant that can be chosen by generalized cross-validation. Our tests on simulated and real data demonstrate the virtue of penalized regression in model building and prediction for panel count data in ultrahigh-dimensional settings.

► Variable selection for high-dimensional panel count data with p>np>n via coordinate ascent.
► Pseudo-objective function based on the estimating function.
► Generalized cross-validation to determine the tuning constant.