Two nonparametric feature screening methods, namely, the Kolmogorov filter and model free, marginally measure the relationship between categorical response and predictor variables without the parametrical assumption. And they can select important variables in the high‐dimensional classification data. Random forest, as a classical nonparametric method, can solve various classification problems. In this paper, we combine the two nonparametric feature screening methods with random forest to handle with spectral data classification. And then other conventional classification methods are compared with ours on three spectral datasets. The comparison results illustrated that our methods have more desirable ability about classification performance and variable selection than other methods.

中文翻译：

---

具有非参数特征筛选的高维光谱数据分类

两种非参数特征筛选方法，即 Kolmogorov 滤波器和无模型，在没有参数假设的情况下略微测量分类响应和预测变量之间的关系。他们可以选择高维分类数据中的重要变量。随机森林作为一种经典的非参数方法，可以解决各种分类问题。在本文中，我们将两种非参数特征筛选方法与随机森林相结合来处理光谱数据分类。然后在三个光谱数据集上将其他传统分类方法与我们的方法进行比较。比较结果表明，我们的方法在分类性能和变量选择方面比其他方法具有更理想的能力。