The article introduces a method for selecting important variables in nonparametric kernel-based estimation. The method involves two stages: first, estimating the target function accurately, and then simplifying the estimator using a small number of variables. This method can be used in various types of nonparametric estimation. The researchers show that when using a specific type of kernel, the method consistently selects the right variables. Experiments demonstrate that this method is effective in both simulated and real-world data scenarios.