Abstract: In car insurance business, to predict the claim frequency become a focus on the research of theory and application in non-life actuarial studies. However, the traditional models and estimation methods can not be applied under the case of high-dimensional information (covariates). Hence, in this paper, based on the mean count model, we first use convex penalty function for variable selection, and then, non-parameter method is employed to get an estimate of the unspecified baseline function. A small simulation and a real data analysis are given as well to assess the proposed model the methods in this paper.